A government affair data intelligent auditing method and configuration system

Abstract

The application discloses a kind of government affair data intelligent auditing method and configuration system, it is related to intelligent decision-making and block chain audit technical field, including, construct federal cognitive twin model, the original business data collected is federated desensitization processing, generates desensitization feature vector set, and input federal cognitive twin model, generates local knowledge graph ciphertext fragment;Local knowledge graph ciphertext fragment is constructed as dynamic space-time fusion graph by homomorphic graph fusion algorithm, and input causal inference model, generates risk transmission probability value;Risk transmission probability value is input into multi-modal risk decision tree, according to the risk transmission threshold discretization into three-level risk level label according to preset, generates hierarchical response instruction set.The application realizes cross-domain data cooperation by federal cognitive twin model, improves the reservation rate of multi-source feature covariance in ciphertext state, solves the feature fusion degradation problem caused by government affair data island.

Description

Technical Field

[0001] This invention relates to the fields of intelligent decision-making and blockchain auditing technology, and in particular to a method and configuration system for intelligent auditing of government data. Background Technology

[0002] Government data auditing technology has evolved from rudimentary systems based on preset rules to semi-automated platforms incorporating machine learning models. In recent years, multi-centralized collaborative architectures have achieved cross-departmental data processing while protecting privacy by introducing secure aggregation and controllable noise mechanisms. Hardware acceleration technology has significantly improved the execution efficiency of complex models, and mainstream solutions combine lightweight anomaly detection algorithms to achieve real-time decision-making. The audit traceability system relies on distributed ledger technology to construct a verifiable chain throughout the entire data processing lifecycle.

[0003] Current intelligent auditing methods for government data have some shortcomings. During the cross-domain alignment of multi-source data, the inherent conflict between privacy protection mechanisms and feature preservation requirements leads to the degradation of the nonlinear coupling relationships of high-dimensional features. This rank loss phenomenon results in the modeling accuracy of entity relationship networks being lower than business requirements, making it unable to support cross-domain dynamic analysis. In addition, the lack of verifiable causal chains in risk transmission mechanisms leads to conflicting counterfactual assumptions in the quantification of intervention effects, causing the calculated risk transmission probability values ​​to deviate from the actual policy effects and potentially leading to misjudgment cases. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent auditing method for government data to solve the problems of feature fusion degradation and spatiotemporal drift in risk transmission.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent auditing method for government data, comprising: constructing a federated cognitive twin model; performing federated desensitization processing on the collected original business data to generate a desensitized feature vector set, and inputting it into the federated cognitive twin model to generate a local knowledge graph ciphertext fragment; constructing the local knowledge graph ciphertext fragment into a dynamic spatiotemporal fusion graph through a homomorphic graph fusion algorithm, and inputting it into a causal inference model to generate a risk transmission probability value; inputting the risk transmission probability value into a multimodal risk decision tree, discretizing it into three-level risk level labels according to a preset risk transmission threshold, and generating a graded response instruction set; synchronizing the graded response instruction set to the government function execution end through a blockchain smart contract to generate a status identification code, generating a chain audit trajectory based on the status identification code, and generating an audit report through a full-link status backtracking method.

[0008] As a preferred embodiment of the intelligent government data review method of the present invention, the specific steps for constructing the federated cognitive twin model are as follows:

[0009] A departmental twin sub-layer is constructed based on the departmental plaintext rule base, a federated aggregation layer is constructed based on homomorphic aggregation operators, and a dynamic mapping layer is constructed based on spatiotemporal semantic alignment mapping rules.

[0010] A federated cognitive twin model is constructed based on the departmental twin sub-layer, the federated aggregation layer, and the dynamic mapping layer.

[0011] As a preferred embodiment of the intelligent review method for government data described in this invention, the specific steps for generating the local knowledge graph encrypted fragment are as follows:

[0012] Collect raw business data from business registration forms, tax returns, and environmental pollution discharge records, and use homomorphic encryption technology to perform federated desensitization processing on sensitive fields of the raw business data to generate a desensitized feature vector set;

[0013] The desensitized feature vector set is input into the federated cognitive twin model, and the federated cognitive twin model is driven by departmental rules to perform encrypted graph generation, generating encrypted fragments of local knowledge graph.

[0014] As a preferred embodiment of the intelligent review method for government data described in this invention, the specific steps for generating the risk transmission probability value are as follows:

[0015] A homomorphic graph fusion algorithm is applied to local knowledge graph ciphertext fragments to generate fully associated ciphertext.

[0016] Injecting spatiotemporal identifiers into fully associated ciphertexts generates dynamic spatiotemporal fusion maps;

[0017] The association layer is defined based on the desensitized feature vector set, the intervention layer is defined based on the fully associated encrypted text, and the counterfactual layer is defined based on the dynamic spatiotemporal fusion graph. A causal inference model is constructed based on the association layer, the intervention layer, and the counterfactual layer.

[0018] The dynamic spatiotemporal fusion map is input into the causal inference model, and risk transmission probability values ​​are generated through hierarchical causal transmission deduction.

[0019] As a preferred embodiment of the intelligent review method for government data described in this invention, the specific steps for generating the hierarchical response instruction set are as follows:

[0020] Define risk transmission thresholds based on the department's plaintext rule base, and calculate default values ​​based on the risk transmission thresholds;

[0021] The effectiveness of the risk transmission probability value is verified by comparing it with the risk transmission threshold.

[0022] If the verification is valid, the data is input into the multimodal risk decision tree, and after the risk transmission threshold discretization classification operation, a three-level risk level label is generated.

[0023] If the validation fails, the default value will be used instead of the invalid value.

[0024] Based on the three-level risk level label, a preset three-level response strategy is triggered to generate a graded response instruction set.

[0025] As a preferred embodiment of the intelligent auditing method for government data described in this invention, the specific steps for generating the chain-like audit trajectory are as follows:

[0026] Invoke the blockchain smart contract to transmit a hierarchical response instruction set and return the transaction hash value; generate a status identifier code based on the transaction hash value.

[0027] An initial root node is created based on the status identifier code, and operational event nodes of the government function execution end are incrementally added on the basis of the initial root node, forming a chain-like audit trajectory by linking them level by level.

[0028] As a preferred embodiment of the intelligent auditing method for government data described in this invention, the method involves: performing reverse traversal of the chain-like audit trajectory using a full-link state backtracking method, collecting the sequence of execution events, structuring it, and generating an audit report.

[0029] Secondly, this invention provides an intelligent auditing system for government data, comprising a federated cognition module, a causal inference module, an instruction hierarchy module, and a report generation module. The federated cognition module is used to construct a federated cognitive twin model, perform federated desensitization processing on the collected raw business data, generate a desensitized feature vector set, and input it into the federated cognitive twin model to generate a local knowledge graph encrypted fragment. The causal inference module is used to construct a dynamic spatiotemporal fusion graph from the local knowledge graph encrypted fragment using a homomorphic graph fusion algorithm, and input it into the causal inference model to generate a risk transmission probability value. The instruction hierarchy module is used to input the risk transmission probability value into a multimodal risk decision tree, discretize it into three levels of risk level labels according to a preset risk transmission threshold, and generate a hierarchical response instruction set. The report generation module is used to synchronize the hierarchical response instruction set to the government function execution end through a blockchain smart contract, generate a status identifier code, generate a chain-like audit trajectory based on the status identifier code, and generate an audit report through a full-link status backtracking method.

[0030] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein the computer program, when executed by the processor, implements any step of the intelligent government data review method as described in the first aspect of the present invention.

[0031] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent government data review method as described in the first aspect of the present invention.

[0032] The beneficial effects of this invention are as follows: cross-domain data collaboration is achieved through a federated cognitive twin model, improving the retention rate of multi-source feature covariance in encrypted form and solving the feature fusion degradation problem caused by government data silos; the effect of policy intervention is dynamically quantified through a causal inference model, and the risk transmission prediction error is compressed by combining spatiotemporal grid identification, solving the spatiotemporal drift defect of risk transmission; and the full lifecycle verifiable traceability of operational events is achieved through blockchain chain audit trajectory, improving audit backtracking efficiency. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart of a method for intelligent review of government data.

[0035] Figure 2 This is a schematic diagram of a government data intelligent review system.

[0036] Figure 3 A flowchart for generating encrypted fragments of a local knowledge graph.

[0037] Figure 4 A flowchart for generating a chain of audit trails. Detailed Implementation

[0038] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0039] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0040] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0041] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for intelligent review of government data, including the following steps:

[0042] S1. Construct a federated cognitive twin model, perform federated desensitization processing on the collected raw business data, generate a desensitized feature vector set, and input it into the federated cognitive twin model to generate local knowledge graph ciphertext fragments.

[0043] S1.1 Construct a department twin sub-layer based on the department plaintext rule base, construct a federated aggregation layer based on the homomorphic aggregation operator, and construct a dynamic mapping layer based on the spatiotemporal semantic alignment mapping rules;

[0044] It should be noted that the construction of the departmental twin layer involves: calling the departmental plaintext rule base (containing business registration verification rules, tax declaration compliance clauses, and environmental pollution discharge thresholds, etc.) through the rule engine API; parsing the conditional logic and parameter constraints in the departmental plaintext rule base; the conditional logic includes "IF-THEN" judgment statements; and the parameter constraints include numerical ranges (such as tax rate ranges of 0-45%), format requirements (such as 18-digit unified social credit code), and logical relationships (such as consistency verification between business registration location and business location); and generating executable departmental rules, including business rules: registered capital threshold verification. Verification (e.g., capital < 3 million triggers micro-enterprise identification) and legal representative consistency verification (matching the legal representative's name in the registered and operating locations); tax rules: income tax reconciliation review (e.g., logical comparison of income and income tax) and tax arrears risk threshold (e.g., cumulative tax arrears > 50,000 triggers an alarm); environmental rules: pollution discharge exceeding standards judgment (e.g., monitoring value > 80mg / L is considered exceeding the standard) and rectification frequency rules (e.g., exceeding the standard more than 3 times in a quarter marks as key supervision), and load the rules of each department into memory to form a departmental twin sub-layer; executable departmental rules include data verification functions, compliance inspection units, and risk warning rules;

[0045] The construction of the federated aggregation layer selects homomorphic aggregation operators based on business scenarios (such as Paillier for financial indicator aggregation and RSA for joint statistics using homomorphic multiplication); it calls the homomorphic encryption library, uses the standard key generation function of the homomorphic encryption library to initialize the operator key pairs, and generates public and private keys; it distributes the public keys to the information nodes of each department, and uses code to call the underlying functions of the homomorphic encryption library to package the homomorphic addition and multiplication operations into standardized API interfaces, generating homomorphic aggregation interfaces and forming the federated aggregation layer;

[0046] The logical relationships of information nodes in each department are encrypted with a public key and input into the federated aggregation layer. An encrypted association matrix is ​​dynamically generated through homomorphic addition or multiplication operations. The construction of the dynamic mapping layer is achieved by parsing spatiotemporal semantic alignment mapping rules (such as mapping environmental grid code 2319 to tax zone C3), extracting the logical relationships of the department twin sub-layers and the encrypted association matrix of the federated aggregation layer. The encrypted spatial computation is performed using a cross-domain entity dynamic linking algorithm: all enterprise nodes in the twin sub-layer are located based on environmental grid code 2319, and the encrypted attributes of all enterprise nodes are attached; all enterprise nodes are mapped to the administrative jurisdiction of tax zone C3 through a spatial coordinate transformation function; finally, an encrypted tree topology structure with tax zone C3 as the parent node and enterprise encrypted entities is dynamically constructed, and spatiotemporal constraints are implanted to form a dynamic mapping layer that supports cross-domain federated computation.

[0047] S1.2 Construct a federated cognitive twin model based on the departmental twin sub-layer, federated aggregation layer, and dynamic mapping layer;

[0048] It should be noted that the departmental rules of the departmental twin sub-layer directly process the raw business data, generating structured business entities and attribute sets; the federated aggregation layer calls the pre-encapsulated homomorphic aggregation interface to perform homomorphic encryption on the structured business entities and attribute sets, generating encrypted entities; the dynamic mapping layer extracts the logical relationships in the encrypted entities, and combines them with predefined environmental grid-tax zone mapping rules, specifically including: spatial responsibility area matching rules specify that environmental grids are divided according to latitude and longitude intervals, and tax zones are divided into different responsibility areas according to administrative boundaries, with each environmental grid precisely corresponding to a single tax zone; entity association rules specify that enterprises belong to environmental grids through their registered addresses, and then map them to the corresponding tax zones, while also supporting reverse indexing; spatiotemporal constraint extension rules dynamically adjust the mapping relationship according to different time periods; if the same environmental grid involves multiple tax zones, conflicts are resolved through the registered address and postal code; and encrypted entities are encrypted and bound together through a cross-domain entity dynamic linking algorithm, completing the construction of the federated cognitive twin model;

[0049] The training of the federated cognitive twin model is achieved through a predefined rule engine-driven, federated homomorphic aggregation update: executable departmental rules loaded in the departmental twin sub-layer are directly applied to the original business data, generating structured entity attributes; the federated aggregation layer encrypts the structured business entities through a homomorphic encryption interface, generating encrypted entities, and performs weighted aggregation operations in the homomorphic ciphertext state; the dynamic mapping layer dynamically adjusts the association weights between encrypted entities based on the real-time input spatiotemporal semantic alignment mapping rules, and dynamically incrementally optimizes the association through a spatiotemporal dynamic association function to complete the training of the federated cognitive twin model. The expression of the federated cognitive twin model is as follows:

[0050] ;

[0051] in, This represents a ciphertext fragment of a local knowledge graph; Represents a spatiotemporal dynamic correlation function; This represents the spatiotemporal semantic alignment mapping rules; This represents the department index, with a value range of 1- ; Indicates the total number of departments; Indicates the first The weight of each department; Represents the homomorphic aggregation operator; Indicates the first The enforceable departmental rules of each department; Indicates the first The original business data of each department.

[0052] S1.3 Collect raw business data from business registration forms, tax returns, and environmental pollution discharge records, and perform federated desensitization processing on sensitive fields of the raw business data through a federated desensitization engine to generate a desensitization feature vector set.

[0053] It should be noted that the data collected includes business registration features from the business registration forms, such as registered capital, registered address, legal person information, and registration timestamp; tax features from tax returns, such as income tax payable, historical tax arrears, and declaration timestamp; and environmental features from environmental discharge records, such as real-time monitoring values, number of exceedances, and original business data with record timestamps. The federated desensitization engine calls the anonymization unit to perform generalization processing on the legal representative's name field, while simultaneously using a Laplace noise injection mechanism to add random perturbations to the numerically sensitive fields. The processed name field and numerically sensitive fields are hashed using SHA3-256 to generate an irreversible federated identifier. The federated identifier, the processed name field, and the numerically sensitive fields are then concatenated along their dimensions to generate a desensitized feature vector set.

[0054] S1.4 Input the desensitized feature vector set into the federated cognitive twin model, and drive the federated cognitive twin model to perform encrypted graph generation through departmental rules to generate local knowledge graph encrypted fragments.

[0055] It should be noted that the desensitized feature vector set is input into the federated cognitive twin model. The executable department rules loaded by the department twin sub-layer process the rule-related fields in the desensitized feature vector set: automatically identify the fields required by the executable department rules, transform the executable department rules into encrypted computation logic in a secure environment, perform encrypted judgment mapping on the target fields to compliance labels, and generate structured business entities; the federated aggregation layer calls the homomorphic aggregation interface to perform homomorphic encryption operation on the sensitive attribute values ​​(business registration capital, tax income and environmental pollution discharge detection values) of the structured business entities to generate encrypted entities; the dynamic mapping layer, based on the spatiotemporal semantic alignment mapping rules, associates the encrypted entities with the corresponding department units through the cross-domain entity dynamic linking algorithm, and finally generates local knowledge graph encrypted fragments including business encrypted fragments, tax encrypted fragments and environmental encrypted fragments with encrypted entities as nodes and cross-domain associations as edges;

[0056] It should also be noted that the business registration encrypted fragment contains encrypted enterprise entity nodes (encrypted attributes of enterprise registered capital and encrypted labels for legal representative identity verification) and jurisdictional association edges (encrypted jurisdictional relationship between the enterprise and its administrative division); the tax encrypted fragment contains encrypted taxpayer entity nodes (encrypted attributes of income tax payable and encrypted labels for tax arrears risk) and tax collection and administration association edges (encrypted attribution relationship between the taxpayer and the tax collection and administration area); the environmental protection encrypted fragment contains encrypted pollution discharge entity nodes (encrypted attributes of real-time pollution discharge concentration and encrypted labels for environmental risk level) and regulatory association edges (encrypted responsibility relationship between the pollution discharge unit and the environmental supervision area).

[0057] S2. Construct a dynamic spatiotemporal fusion graph from local knowledge graph ciphertext fragments using a homomorphic graph fusion algorithm, and input it into a causal inference model to generate risk transmission probability values;

[0058] S2.1 Execute the homomorphic graph fusion algorithm on the local knowledge graph ciphertext fragments to generate fully associated ciphertext;

[0059] It should be noted that the homomorphic graph fusion algorithm is applied to the local knowledge graph ciphertext fragments to parse the cross-domain association edges of multiple ciphertext fragments, and the ciphertext attributes of the same encrypted entities are fused based on the homomorphic addition operator; the weights of the cross-domain association edges of the cross-domain fragments are encrypted and superimposed using homomorphic multiply-addition operations to generate fully associated ciphertext containing encrypted entity nodes and cross-domain association edges.

[0060] S2.2 Inject spatiotemporal identifiers into the fully associated ciphertext to generate a dynamic spatiotemporal fusion graph;

[0061] It should be noted that the registration address in the fully associated ciphertext is extracted through spatiotemporal semantic alignment mapping rules. The registration coordinates are converted into latitude and longitude coordinates through the government geocoding service interface, and the output address string is the coordinate value. A jurisdiction mapping rule library is defined based on the digital city geocoding standard, which includes entries for coordinate values ​​and responsibility area codes. The jurisdiction mapping rule library is called according to the coordinate values ​​to generate the corresponding responsibility area code. The timestamp ciphertext in each encrypted entity is collected and homomorphic decryption is performed to output a 13-bit plaintext millisecond timestamp. The responsibility area code and the millisecond timestamp are used as spatiotemporal identifiers and injected into the tail of the encrypted entity node and the cross-domain association edge through string concatenation to generate a dynamic spatiotemporal fusion graph.

[0062] S2.3. Define the association layer based on the desensitized feature vector set, the intervention layer based on the fully associated encrypted text, and the counterfactual layer based on the dynamic spatiotemporal fusion graph. Construct a causal inference model based on the association layer, the intervention layer, and the counterfactual layer.

[0063] It should be noted that the covariance matrix framework is defined based on the desensitized feature vector set. That is, an empty ciphertext matrix is ​​created, and all matrix elements in each covariance matrix are preset to the initial value of homomorphic encryption zero value. The row and column indices of the covariance matrix correspond to each desensitized feature vector in the desensitized feature vector set. Homomorphic multiplication, addition and subtraction operators are configured to perform homomorphic multiplication on the input desensitized feature vector set to calculate the desensitized feature product. Then, homomorphic addition is used to accumulate the desensitized feature product. Finally, homomorphic subtraction is used to subtract the mean desensitized feature, establish the covariance calculation protocol between desensitized features, and complete the construction of the association layer.

[0064] Based on fully associated ciphertext encrypted entity nodes and cross-domain association edges, a dual-buffered structure for intervention and control groups is defined: two independent storage containers are constructed, one for the intervention group to store encrypted entities to be processed and the other for the control group to store encrypted entities that do not need to be processed, both stored in hash table form; a homomorphic subtraction operator is loaded to calculate the mean of the ciphertext risk values ​​for the intervention and control groups respectively, and the difference between the means of the ciphertext risk values ​​of the two groups is calculated by the homomorphic subtraction operator, and the difference is encrypted and output as a fixed-length ciphertext, an average intervention effect calculation protocol is established, and the empty framework of the intervention layer is completed;

[0065] Based on the dynamic spatiotemporal fusion graph, attribute modification slots and risk recalculation output slots are defined: the attribute modification slot is defined as a plaintext numerical input port to receive the attribute value to be modified, and the risk recalculation output slot is defined as a plaintext numerical output port to return the counterfactual risk value; based on the responsibility area code and millisecond timestamp in the dynamic spatiotemporal fusion graph, executable department rules for the corresponding region and time period are extracted; a counterfactual inference protocol is established: the applicable business rule version is determined through the executable department rules, the attribute value to be modified in the attribute modification slot is set as the hypothetical value, the executable department rules for the corresponding region and time period are called to calculate the risk change value, and the counterfactual risk value is calculated based on the hypothetical value and the risk change value and output to the risk recalculation output slot; the empty framework of the counterfactual layer is completed; a causal inference model is constructed based on the association layer, intervention layer, and counterfactual layer;

[0066] The training of the causal inference model is as follows: The association layer pre-configures a covariance calculation protocol between desensitized features (verifying the numerical precision error of homomorphic multiplication, addition, and subtraction operators to less than one part per million using historical desensitized feature vector sets); the intervention layer is divided into intervention and control groups, with thousands of encrypted test entities injected to verify a grouping accuracy exceeding 99.9%; the counterfactual layer solidifies the applicable business rule version; the three-level framework is integrated and deployed after independent stress testing; subsequent support will only include replacement of applicable business rule versions (e.g., a full reload of the rule library when new environmental regulations are released); the expression is...

[0067] ;

[0068] in, This represents the probability value of risk transmission. Represents the homomorphic graph fusion operator; Indicates the encrypted entity of the intervention group; This represents the encrypted entity of the control group; This represents the ciphertext mean of the risk values ​​for the intervention group; This represents the ciphertext mean of the risk value for the control group. This indicates the weight parameter of the applicable business rule version.

[0069] S2.4 Input the dynamic spatiotemporal fusion map into the causal inference model, and generate risk transmission probability values ​​through hierarchical causal transmission deduction.

[0070] It should be noted that the dynamic spatiotemporal fusion graph is input into the causal inference model. The desensitized feature vector set of encrypted entity nodes in the graph is extracted at the correlation layer. A ciphertext correlation identifier is generated by executing a covariance calculation protocol between desensitized features, and the ciphertext correlation identifier is homomorphically decrypted to generate a decrypted value. Intervention conditions are defined based on the ciphertext correlation identifier, including intervention variables and intervention thresholds. A baseline value of 0.5 is defined according to the rules of government risk transmission. When the decrypted value is greater than the baseline value, the intervention variable is set as a numerical feature in the desensitized feature vector set. Homomorphic percentile calculation is performed on the intervention variable. The process involves generating a baseline distribution value for the intervention variable, and then generating an intervention threshold based on the baseline distribution value and the decrypted value. The intervention group and control group are divided according to the intervention conditions: the encrypted value of the intervention variable is parsed using a homomorphic comparison operator; encrypted entities whose encrypted value is greater than the intervention threshold are assigned to the intervention group, and those whose encrypted value is less than or equal to the intervention threshold are assigned to the control group. The encrypted intervention effect value is output by executing an average intervention effect calculation protocol. Counterfactual inference parameters are defined based on the encrypted intervention effect value, and a counterfactual inference protocol is executed based on these parameters to output the risk transmission probability value.

[0071] S3. Input the risk transmission probability value into the multimodal risk decision tree, discretize it into three-level risk level labels according to the preset risk transmission threshold, and generate a hierarchical response instruction set;

[0072] S3.1 Define risk transmission thresholds based on the department's explicit rule base, and calculate default values ​​based on the risk transmission thresholds;

[0073] It should be noted that the risk transmission threshold includes a low-risk threshold and a high-risk threshold; the low-risk threshold and high-risk threshold fields in the department's plaintext rule base are parsed, and the corresponding values ​​of 0.3 and 0.7 bound to the fields are matched by the rule parsing engine; the arithmetic average of the low-risk threshold and the high-risk threshold is calculated to generate a default value;

[0074] It should also be noted that, according to the risk classification and control specifications, the low-risk threshold is 0.3 and the high-risk threshold is 0.7. Based on a five-year cross-departmental event database, statistical analysis (over 100,000 cases) shows that when the risk transmission probability value is ≤0.3, the actual transmission rate is <5%, and when the risk transmission probability value is >0.7, the event will trigger transmission 100%.

[0075] S3.2 Verify the validity of the risk transmission probability value. If valid, input it directly into the multimodal risk decision tree. If invalid, use the default value to replace the invalid value.

[0076] It should be noted that when obtaining the risk transmission probability value, it is necessary to determine whether it is a numerical type and whether it is within the closed interval [0, 1], i.e., ≥0 and ≤1. If the risk transmission probability value is a numerical type and is within the closed interval [0, 1], then the risk transmission probability value is directly retained and input into the multimodal risk decision tree. If either condition is violated, then the default value is used to replace the risk transmission probability value and input into the multimodal risk decision tree.

[0077] S3.3 Call the multimodal risk decision tree, input the risk transmission probability value into the multimodal risk decision tree, and generate three-level risk level labels after the risk transmission threshold discretization classification operation;

[0078] It should be noted that the framework of the multimodal risk decision tree is a three-level classification rule engine. The risk transmission thresholds (0.3 and 0.7) are hard-coded into conditional judgment functions. After encoding, the conditional judgment functions can be directly called and the risk transmission probability values ​​can be input to generate three-level risk level labels. The risk transmission probability values ​​are input into the multimodal risk decision tree to perform discretization classification: if the risk transmission probability value is <0.3, the label "Low" is output; if the risk transmission probability value is ≥0.3 and <0.7, the label "Medium" is output; if the risk transmission probability value is ≥0.7, the label "High" is output. Finally, the three-level risk level labels in plaintext form are generated.

[0079] S3.4. Trigger the preset three-level response strategy based on the three-level risk level label to generate a graded response instruction set.

[0080] It should be noted that, through key-value matching rules: "Low" is mapped to "basic monitoring and periodic reporting", "Medium" is mapped to "enhanced monitoring and contingency plan activation", and "High" is mapped to "emergency intervention and interdepartmental coordination", the corresponding level 3 response strategy text is triggered according to the level 3 risk level label; the level 3 response strategy matched by the current level 3 risk level label is filled into the corresponding key value in the graded response instruction set (e.g., if the level 3 risk level label is "High", then "emergency intervention and interdepartmental coordination" is filled into the High key), and other key positions are left blank, and the graded response instruction set is output.

[0081] S4. Synchronize the hierarchical response instruction set to the government function execution end through blockchain smart contracts, generate status identification codes, generate chain audit traces based on status identification codes, and generate audit reports through full-link status backtracking methods.

[0082] S4.1, Invoke the blockchain smart contract to transmit the hierarchical response instruction set and return the transaction hash value, and generate a status identifier code based on the transaction hash value;

[0083] It should be noted that the construction process of blockchain smart contracts includes five steps: solidification of government needs, contract development, testing and verification, auditing and hardening, and mainnet deployment. The government departments clarify business rules (such as instruction set synchronization trigger conditions and government function execution terminal permissions) and solidify them into technical documents. Smart contract code is written using the Solidity language to implement hierarchical response instruction set receiving functions and transaction hash generation logic. Full-process testing is simulated on a private test chain. Risks such as reentrancy vulnerabilities are detected and fixed using the security auditing tool Slither. The code is compiled into EVM bytecode and deployed to the government consortium blockchain, with the call channel opened through the ABI interface.

[0084] The blockchain smart contract's ABI interface is invoked to transmit a tiered response instruction set and generate a transaction request containing the tiered response instruction content and a millisecond timestamp. The confirmation process of the blockchain smart contract includes: the government function execution end calls the ABI interface to send a signed transaction carrying the tiered response instruction set and a millisecond timestamp; after the signed transaction is verified by the blockchain node for its validity, instruction format compliance, and the validity of the verification node's transaction, it is broadcast to the entire network; the blockchain node packages it into a candidate block through Byzantine fault-tolerant consensus; the candidate blockchain is connected to the main chain; the blockchain smart contract automatically generates a unique transaction hash value and returns it to the government function execution end; the transaction hash value, millisecond timestamp, and responsibility area code are combined to generate a raw string; and a SHA-256 hash operation is performed on the raw string to generate a fixed-length 64-character status identifier code.

[0085] S4.2. Create an initial root node based on the status identifier code, and incrementally add operation event nodes of the government function execution end on the basis of the initial root node, linking them level by level to form a chain audit trajectory;

[0086] It should be noted that the initial root node content is based on the status identifier code, and a millisecond timestamp is added as the creation timestamp of the initial root node to generate a unique hash identifier for the initial root node. When the government function execution end performs an operation event (such as instruction parsing and cross-departmental coordination), the operation type (such as "instruction parsing"), the execution timestamp of the operation at the government function execution end, and the hash identifier of the previous event node are extracted to construct the content of the new event node. The hash identifier of the new event node content is calculated as the new node identifier. Subsequent operations repeatedly create nodes, each time referencing the hash representation of the previous new event node as the connection basis, forming a hierarchical hash reference chain of "initial root node → new event node 1 → new event node 2", and generating the execution conclusion (success, timeout, and failure) in the new event node at the end, thus constructing a chain-like audit trajectory.

[0087] S4.3. The chain audit trajectory is traversed in reverse through the full-link state backtracking method to collect event sequences and structure them into an audit report.

[0088] It should be noted that, based on the hash identifier of the last new event node in the chain-like audit trajectory, the chain-like audit trajectory is traversed in reverse through node pointers: the operation type, execution timestamp of the operation at the government function execution end, and hash identifier of the previous new event node are extracted from the content of the last new event node and stored in the event temporary storage queue; the hash identifier of the previous new event node is used as an index (HN-1) to locate the previous event node (N-1), the content is parsed and stored in the event temporary storage queue and the index value is updated (HN-2); the location-parsing-update process is repeated until the initial root node is traced back, and the status identifier code and creation timestamp in the content of the initial root node are parsed as the first item in the event temporary storage queue; the new event nodes in the event temporary storage queue are sorted in ascending order of execution timestamp (the root node creation timestamp is the earliest, and the last new event node execution timestamp is the latest), and structurally assembled into an audit report, which includes a unique identifier for the report (i.e., status identifier code), the timeline of the new event node, and the execution conclusion (operation status of the last new event node), thus completing the generation of the audit report.

[0089] This embodiment also provides an intelligent government data review system, including: a federal cognitive module, a causal inference module, an instruction classification module, and a report generation module;

[0090] The Federated Cognition Module is used to build a Federated Cognitive Twin Model. It performs federated desensitization processing on the collected raw business data, generates a desensitized feature vector set, and inputs it into the Federated Cognitive Twin Model to generate local knowledge graph encrypted fragments.

[0091] The causal inference module is used to construct a dynamic spatiotemporal fusion graph from local knowledge graph ciphertext fragments using a homomorphic graph fusion algorithm, and input it into the causal inference model to generate risk transmission probability values.

[0092] The instruction grading module is used to input the risk transmission probability value into the multimodal risk decision tree, discretize it into three-level risk level labels according to the preset risk transmission threshold, and generate a graded response instruction set.

[0093] The report generation module is used to synchronize the hierarchical response instruction set to the government function execution end through blockchain smart contracts, generate status identification codes, generate chain audit traces based on status identification codes, and generate audit reports through the full-link status backtracking method.

[0094] This embodiment also provides a computer device applicable to the intelligent review method for government data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the intelligent review method for government data as proposed in the above embodiment.

[0095] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0096] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the intelligent review method for government data as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage 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 Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0097] In summary, this invention achieves cross-domain data collaboration through: a federated cognitive twin model, improving the retention rate of multi-source feature covariance in encrypted form, and solving the feature fusion degradation problem caused by siloed government data; dynamically quantifies the policy intervention effect through a causal inference model, and compresses risk transmission prediction errors by combining spatiotemporal grid identification, thus solving the spatiotemporal drift defect of risk transmission; and achieves full lifecycle verifiable traceability of operational events through blockchain chain audit trajectories, improving audit backtracking efficiency.

[0098] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent review of government data, characterized in that: include, Construct a federated cognitive twin model, perform federated desensitization processing on the collected raw business data, generate a desensitized feature vector set, and input it into the federated cognitive twin model to generate local knowledge graph ciphertext fragments; The local knowledge graph ciphertext fragments are constructed into a dynamic spatiotemporal fusion graph using a homomorphic graph fusion algorithm, and then input into a causal inference model to generate risk transmission probability values. The risk transmission probability value is input into the multimodal risk decision tree, and discretized into three-level risk level labels according to the preset risk transmission threshold to generate a hierarchical response instruction set; The hierarchical response instruction set is synchronized to the government function execution end through blockchain smart contracts, a status identification code is generated, a chain audit track is generated based on the status identification code, and an audit report is generated through the full-link status backtracking method.

2. The intelligent review method for government data as described in claim 1, characterized in that: The specific steps for constructing the federated cognitive twin model are as follows: A departmental twin sub-layer is constructed based on the departmental plaintext rule base, a federated aggregation layer is constructed based on homomorphic aggregation operators, and a dynamic mapping layer is constructed based on spatiotemporal semantic alignment mapping rules. A federated cognitive twin model is constructed based on the departmental twin sub-layer, the federated aggregation layer, and the dynamic mapping layer.

3. The intelligent review method for government data as described in claim 2, characterized in that: The specific steps for generating encrypted fragments of the local knowledge graph are as follows. Collect raw business data from business registration forms, tax returns, and environmental pollution discharge records, and use homomorphic encryption technology to perform federated desensitization processing on sensitive fields of the raw business data to generate a desensitized feature vector set; The desensitized feature vector set is input into the federated cognitive twin model, and the federated cognitive twin model is driven by departmental rules to perform encrypted graph generation, generating encrypted fragments of local knowledge graph.

4. The intelligent review method for government data as described in claim 3, characterized in that: The specific steps for generating the risk transmission probability value are as follows: A homomorphic graph fusion algorithm is applied to local knowledge graph ciphertext fragments to generate fully associated ciphertext. Injecting spatiotemporal identifiers into fully associated ciphertexts generates dynamic spatiotemporal fusion maps; The association layer is defined based on the desensitized feature vector set, the intervention layer is defined based on the fully associated encrypted text, and the counterfactual layer is defined based on the dynamic spatiotemporal fusion graph. A causal inference model is constructed based on the association layer, the intervention layer, and the counterfactual layer. The dynamic spatiotemporal fusion map is input into the causal inference model, and risk transmission probability values ​​are generated through hierarchical causal transmission deduction.

5. The intelligent review method for government data as described in claim 4, characterized in that: The specific steps for generating the hierarchical response instruction set are as follows. Define risk transmission thresholds based on the department's plaintext rule base, and calculate default values ​​based on the risk transmission thresholds; The effectiveness of the risk transmission probability value is verified by comparing it with the risk transmission threshold. If the verification is valid, the data is input into the multimodal risk decision tree, and after the risk transmission threshold discretization classification operation, a three-level risk level label is generated. If the validation fails, the default value will be used instead of the invalid value. Based on the three-level risk level label, a preset three-level response strategy is triggered to generate a graded response instruction set.

6. The intelligent review method for government data as described in claim 5, characterized in that: The specific steps for generating the chain-like audit trail are as follows. Invoke the blockchain smart contract to transmit a hierarchical response instruction set and return the transaction hash value; generate a status identifier code based on the transaction hash value. An initial root node is created based on the status identifier code, and operational event nodes of the government function execution end are incrementally added on the basis of the initial root node, forming a chain-like audit trajectory by linking them level by level.

7. The intelligent review method for government data as described in claim 6, characterized in that: The method involves reverse traversal of the chain audit trajectory using the full-link state backtracking method, collecting the sequence of execution events, structuring it, and generating an audit report.

8. A government data intelligent review system, based on the government data intelligent review method according to any one of claims 1 to 7, characterized in that: This includes the federal cognitive module, the causal inference module, the instruction hierarchy module, and the report generation module; The Federated Cognition Module is used to build a Federated Cognitive Twin Model. It performs federated desensitization processing on the collected raw business data, generates a desensitized feature vector set, and inputs it into the Federated Cognitive Twin Model to generate local knowledge graph encrypted fragments. The causal inference module is used to construct a dynamic spatiotemporal fusion graph from local knowledge graph ciphertext fragments using a homomorphic graph fusion algorithm, and input it into the causal inference model to generate risk transmission probability values. The instruction grading module is used to input the risk transmission probability value into the multimodal risk decision tree, discretize it into three-level risk level labels according to the preset risk transmission threshold, and generate a graded response instruction set. The report generation module is used to synchronize the hierarchical response instruction set to the government function execution end through blockchain smart contracts, generate status identification codes, generate chain audit traces based on status identification codes, and generate audit reports through the full-link status backtracking method.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent government data review method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent auditing method for government data as described in any one of claims 1 to 7.

Images