Government affair inspection method and system based on information management and natural language processing

Abstract

The invention discloses a government affair inspection method and system based on information management and natural language processing, and belongs to the technical field of big data analysis and natural language process.The method comprises the steps that a micro-service architecture system is adopted to construct a government affair inspection information management platform, and an inspection group is supported to access the system in real time through multiple terminals; the request data is transmitted to a security domain server through a government affair external network after being encrypted by a national cryptographic algorithm; an intelligent screening mechanism based on natural language processing and Transform model driving; performing hierarchical data security protection: establishing an authority control model, and dividing a data access domain according to roles; performing problem qualitative analysis based on natural language processing and Transform model driving; and analyzing the patrol data and generating an intelligent report. According to the method, full-process digitalization, full-factor association and full-period intelligentization of patrol work can be realized, and the problem discovery accuracy, the analysis depth and the rectification effectiveness are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis and natural language processing, and specifically to a government inspection method and system based on information management and natural language processing. Background Art

[0002] While a systematic framework has been established for government inspections, significant shortcomings remain in its information technology support capabilities. Existing inspection information systems generally lag behind the demands of digital governance and lack a standardized, integrated technical framework. This makes it difficult to adapt to the dynamic and complex nature of government oversight under the new circumstances.

[0003] Government inspections involve the integration and analysis of multi-dimensional data resources, including organizational structures, staffing information, functional allocation lists, fiscal budget implementation reports, minutes of major decision-making meetings, cadre appointment and removal documents, special work plans, key position performance records, state-owned asset change ledgers, and internal control systems. This data is often isolated and fragmented, stored in diverse formats, and lacks relevance. This leads to significant challenges during inspections, such as inefficient information retrieval, insufficient data mining depth, and difficulties in cross-departmental collaboration. Existing inspection methods leave significant room for improvement: inspection strategies lack dynamic adaptability and fail to fully integrate the functional characteristics, risk profiles, and historical rectification history of inspected units to create differentiated inspection models. The professional competence of inspectors varies widely, and their understanding of policies and regulations and familiarity with business systems need improvement. The inspection implementation process tends to be formulaic, with limited on-site verification methods and insufficient use of intelligent analytical tools. This limits problem detection and reduces supervisory effectiveness.

[0004] Furthermore, the closed-loop management mechanism for inspection and rectification urgently needs improvement: The implementation of responsibility for problem rectification is characterized by a "hot at the top, lukewarm in the middle, and cold at the bottom" phenomenon. Rectification measures lack rigid constraints, and there are no quantitative standards for evaluating rectification effectiveness, resulting in a low conversion rate of inspection results. Some units tend to prioritize inspection over rectification, with rectification plans becoming superficial and implementation lacking follow-up and accountability, creating a governance dilemma of "discovery-rectification-rebound."

[0005] In order to break through the above-mentioned technical bottlenecks, it is urgent to build a new generation of intelligent patrol technology system. Summary of the Invention

[0006] The technical task of the present invention is to address the above shortcomings and provide a government inspection method and system based on information management and natural language processing, which can realize the full-process digitization, full-factor correlation and full-cycle intelligence of the inspection work, and significantly improve the accuracy of problem discovery, analysis depth and rectification effectiveness.

[0007] The technical solution adopted by the present invention to solve its technical problem is:

[0008] A government inspection method based on information management and natural language processing, the implementation of which includes:

[0009] A microservices architecture is used to build a government inspection information management platform, supporting inspection teams to access the system in real time through multiple terminals (including dedicated terminals). Requested data is encrypted using a national secret algorithm and then transmitted to a secure domain server via the government extranet. Standardized data interfaces are implemented with 12 government systems, including the disciplinary inspection and supervision system, the organization and personnel system, and the financial supervision system, establishing a dynamic mapping mechanism across the three dimensions of "data-business-process."

[0010] An intelligent screening mechanism driven by natural language processing and the Transformer model, including: construction of a knowledge base of inspection observation points, multimodal document parsing, intelligent screening, and evidence chain construction;

[0011] Hierarchical data security protection: Establish a permission control model and divide data access domains according to roles;

[0012] Qualitative problem analysis based on natural language processing and Transformer model-driven, including: multi-level semantic parsing, intelligent comparison of typical cases, and intelligent decision-making support;

[0013] Analyze inspection data and generate intelligent reports: establish a three-level analysis system of "data-model-report"; use knowledge graph technology to build an inspection problem association network, and automatically generate inspection analysis reports including graded risk warnings; support multi-dimensional data drilling function, and conduct interactive analysis according to multiple dimensions.

[0014] This method realizes the full-process digitization, full-factor correlation and full-cycle intelligence of inspection work by constructing a multi-source data fusion model, an intelligent semantic analysis engine and a dynamic risk warning mechanism, significantly improving the accuracy of problem discovery, the depth of analysis and the effectiveness of rectification.

[0015] Furthermore, the inspection observation point knowledge base is constructed:

[0016] Integrate a database of typical inspection cases from the past five years and use the improved Transformer-XL model for long text pre-training;

[0017] A semantic feature vector library containing 128 types of typical problems is constructed, and metadata including risk levels, domain labels, etc. are annotated.

[0018] Furthermore, the multimodal document parsing,

[0019] Adopting OCR+NLP dual-engine architecture, it performs layout analysis, table recognition and semantic parsing on scanned documents;

[0020] Combined with the domain-adaptive Transformer model, accurate parsing of vertical field texts including government terminology, policies and regulations can be achieved.

[0021] Furthermore, the intelligent screening and evidence chain construction,

[0022] The Transformer model based on the multi-head attention mechanism realizes cross-document information association;

[0023] Automatically generate a problem evidence chain map, supporting tracing the problem back to the original document page number.

[0024] Furthermore, the hierarchical data security protection,

[0025] Federated learning technology is used to achieve cross-departmental collaborative data analysis, and problem clue data is transmitted using zero-knowledge proof encryption; a data sandbox mechanism is implemented to ensure that problem data is processed in a closed environment.

[0026] Furthermore, the permission control model is a hybrid permission model based on RBAC+ABAC, which is implemented through a permission inheritance algorithm based on the organizational structure, and establishes a permission audit log analysis model based on temporary permission allocation and recovery.

[0027] Furthermore, the qualitative analysis of the problem driven by natural language processing and Transformer model is

[0028] The multi-level semantic analysis:

[0029] Use the BERT-CRF model for entity recognition and relationship extraction;

[0030] The Transformer model based on the hierarchical attention mechanism realizes long text encoding;

[0031] Intelligent comparison of typical cases:

[0032] Build a semantic retrieval library covering 3,000+ typical cases;

[0033] Adopt multi-granularity semantic similarity calculation method to improve comparison accuracy;

[0034] The intelligent auxiliary decision-making:

[0035] Automatically generate qualitative suggestion options for the problem based on similarity calculation results;

[0036] Supports users to customize the comparison threshold and recommendation result sorting strategy.

[0037] The present invention also claims protection for a government inspection system based on information management and natural language processing, comprising:

[0038] Government Inspection Information Management Platform: This platform utilizes a microservices architecture to build a government inspection information management platform. This platform supports real-time access to the system by inspection teams through multiple terminals (including dedicated terminals). Requested data is encrypted using a national secret algorithm and then transmitted to a secure domain server via the government extranet. It also implements standardized data interfaces with 12 government systems, including the disciplinary inspection and supervision system, the organizational personnel system, and the financial supervision system, establishing a dynamic mapping mechanism across the three dimensions of "data-business-process."

[0039] An intelligent screening module based on natural language processing and Transformer model-driven, including: construction of a knowledge base of inspection observation points, multimodal document parsing, intelligent screening and evidence chain construction;

[0040] Hierarchical data security protection module, used to establish permission control models and divide data access domains according to roles;

[0041] Qualitative problem analysis module, which is driven by natural language processing and the Transformer model, includes: multi-level semantic parsing, intelligent comparison of typical cases, and intelligent decision-making assistance;

[0042] The inspection data analysis and intelligent report generation module establishes a three-level analysis system of "data-model-report", uses knowledge graph technology to construct an inspection problem association network, and automatically generates an inspection analysis report including graded risk warnings;

[0043] The system implements government inspection work through the above method.

[0044] The present invention also claims protection for a device for implementing government inspections based on information management and natural language processing, comprising: at least one memory and at least one processor;

[0045] The at least one memory is configured to store a machine-readable program;

[0046] The at least one processor is configured to call the machine-readable program to implement the above method.

[0047] The present invention also claims protection for a computer-readable medium having computer instructions stored thereon, which are capable of implementing the above method when executed by a processor.

[0048] Compared with the prior art, the government inspection method and system based on information management and natural language processing of the present invention have the following beneficial effects:

[0049] 1. Improved precision in problem identification. By building a multimodal semantic analysis engine based on Transformer-XL, we achieve intelligent screening and in-depth analysis of government documents. The system can automatically identify 128 typical problem characteristics, increasing problem detection accuracy by over 60% compared to traditional manual methods. This effectively addresses the "needle in a haystack" problem faced during government inspections and significantly reduces missed reports and false positives.

[0050] 2. Construction of an intelligent decision support system. The integrated multi-granularity semantic similarity calculation module enables in-depth analysis of the essence of a problem at three levels (appearance, association, and root cause). By linking over 5,000 policies and regulations and over 2,000 typical cases through a knowledge graph, this system automates and accurately characterizes problems, increasing problem characterization efficiency by three times and achieving an adoption rate of recommended solutions exceeding 85%.

[0051] 3. Empowering digital collaboration throughout the entire process. We've created a three-in-one intelligent inspection platform covering pre-, mid-, and post-inspection tasks, automating 23 core processes, including task allocation, progress monitoring, and report generation. Leveraging RPA process robotics technology, we've reduced manual operations by 40%, shortening inspection cycles by an average of 25%, and effectively improving the responsiveness and efficiency of government inspections.

[0052] 4. Strengthening the transparent governance system. Establish a full-process electronic archive system to ensure operational traceability, data traceability, and audit tracking. The system supports 10 years of data storage and offers multi-dimensional retrieval and analysis capabilities, providing verifiable data support for disciplinary inspection, supervision, and audit oversight. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flowchart of a system using NLP and Transformer models to assist in problem screening, provided by one embodiment of the present invention;

[0054] Figure 2 This is a flowchart of a system provided by an embodiment of the present invention using NLP and Transformer models to assist in qualitative analysis of problems. DETAILED DESCRIPTION

[0055] The present invention will be further described below with reference to specific embodiments.

[0056] An embodiment of the present invention provides a government inspection method based on information management and natural language processing. By introducing an improved Transformer model and combining it with a multimodal NLP technology framework, this method constructs an intelligent inspection information processing platform with high robustness and strong generalization capability, thereby improving the ability to accurately identify, deeply trace and systematically govern complex problems.

[0057] The implementation of this method includes:

[0058] 1. Build a government inspection information management platform.

[0059] A microservice architecture system is used to build a government inspection information management platform, which supports the inspection team to access the system in real time through multiple terminals (including dedicated terminals). The requested data is encrypted with the national secret algorithm and then transmitted to the security domain server through the government external network.

[0060] Realize standardized data interfaces with 12 types of government systems, including the discipline inspection and supervision system, organization and personnel system, and financial supervision system, and establish a three-dimensional dynamic mapping mechanism of "data-business-process".

[0061] 2. Intelligent screening mechanism driven by natural language processing (NLP) and Transformer model.

[0062] refer to Figure 1 As shown in the figure, the intelligent screening mechanism driven by natural language processing and Transformer model includes: construction of inspection observation point knowledge base, multimodal document parsing, intelligent screening and evidence chain construction.

[0063] Construction of the knowledge base of inspection observation points:

[0064] We integrated a database of typical central and local inspection cases from the past five years and used the improved Transformer-XL model for long text pre-training.

[0065] Build a semantic feature vector library containing 128 types of typical problems, annotating metadata such as risk level and domain label.

[0066] Multimodal document parsing:

[0067] It adopts OCR+NLP dual-engine architecture to perform layout analysis, table recognition and semantic parsing on scanned documents.

[0068] Combined with the domain-adaptive Transformer model, accurate parsing of vertical texts such as government terminology, policies and regulations can be achieved.

[0069] Intelligent screening and evidence chain construction:

[0070] The Transformer model based on the multi-head attention mechanism realizes cross-document information association.

[0071] Automatically generate a problem evidence chain map, supporting tracing the problem back to the original document page number.

[0072] 3. Hierarchical data security protection:

[0073] A "five-level, three-domain" permission control model was established, dividing data access domains based on roles such as inspection teams, functional departments, and inspected units. Federated learning technology was used to enable cross-departmental data collaborative analysis, and problem clue data was encrypted and transmitted using zero-knowledge proofs. A data sandbox mechanism was implemented to ensure that problematic data was processed in a closed environment.

[0074] 4. Qualitative analysis of problems driven by natural language processing (NLP) and Transformer models:

[0075] refer to Figure 2 As shown in the figure, the qualitative analysis of problems driven by natural language processing and Transformer models includes: multi-level semantic parsing, intelligent comparison of typical cases, and intelligent decision-making assistance.

[0076] Multi-level semantic analysis:

[0077] Use the BERT-CRF model for entity recognition and relationship extraction.

[0078] Transformer model based on hierarchical attention mechanism realizes long text encoding.

[0079] Intelligent comparison of typical cases:

[0080] Build a semantic retrieval library covering 3,000+ typical cases.

[0081] A multi-granularity semantic similarity calculation method is used to improve the comparison accuracy.

[0082] Intelligent decision-making support:

[0083] Automatically generate qualitative suggestion options for the problem based on the similarity calculation results.

[0084] Supports users to customize the comparison threshold and recommendation result sorting strategy.

[0085] 5. Analyze inspection data and generate intelligent reports:

[0086] A three-tiered analysis system, "data-model-report," has been established, covering 18 core indicators, including petition reports, problem leads, and rectification effectiveness. Knowledge graph technology is used to construct a correlation network for inspection issues, automatically generating inspection analysis reports with five levels of risk warnings. Multi-dimensional data drilling is supported, enabling interactive analysis across over 20 dimensions, including time, unit, and issue type.

[0087] This breakthrough approach introduces the Transformer-XL model to the field of government inspections, building a multimodal semantic analysis system with a self-attention enhancement mechanism. Through hierarchical feature extraction (word-level, sentence-level, and paragraph-level) and dynamic context modeling, this engine achieves: an 80% increase in intelligent document screening efficiency, a 92% accuracy rate for problem detection, and the expansion of deep semantic understanding to six dimensions, including policy relevance and risk transmission pathways.

[0088] The innovative adoption of a microservice B / S architecture system achieves: cross-platform access response time <1.2 seconds, system scalability supports tens of thousands of concurrent users, and modular design supports dynamic loading of 12 functional components such as policy and regulations library, case library, and knowledge graph.

[0089] Build a hybrid permission model based on RBAC+ABAC to achieve: permission granularity is refined to the operation level (such as view / edit / export), support dynamic configuration of permissions for 32 roles and positions, and 100% completeness of data access audit logs.

[0090] Develop an inspection process automation system based on a workflow engine, achieving: 85% of inspection links are executed automatically, 95% accuracy rate of abnormal warning of process nodes, and 40% improvement in cross-departmental collaboration efficiency.

[0091] An embodiment of the present invention also provides a government inspection system based on information management and natural language processing, which implements government inspection work through the government inspection method based on information management and natural language processing described in the above embodiment.

[0092] The system includes:

[0093] 1. Government Inspection Information Management Platform: A microservices architecture is used to build a government inspection information management platform, which supports inspection teams to access the system in real time through multiple terminals (including dedicated terminals). The requested data is encrypted with the national secret algorithm and transmitted to the security domain server through the government external network. It realizes standardized data interfaces with 12 types of government systems such as the discipline inspection and supervision system, organization and personnel system, and financial supervision system, and establishes a three-dimensional dynamic mapping mechanism of "data-business-process".

[0094] 2. Intelligent screening module based on natural language processing and Transformer model drive, including: construction of inspection observation point knowledge base, multimodal document parsing, intelligent screening and evidence chain construction.

[0095] The inspection observation point knowledge base is constructed as follows:

[0096] We integrated a database of typical central and local inspection cases from the past five years and used the improved Transformer-XL model for long text pre-training.

[0097] Build a semantic feature vector library containing 128 types of typical problems, annotating metadata such as risk level and domain label.

[0098] The multimodal document parsing:

[0099] It adopts OCR+NLP dual-engine architecture to perform layout analysis, table recognition and semantic parsing on scanned documents.

[0100] Combined with the domain-adaptive Transformer model, accurate parsing of vertical texts such as government terminology, policies and regulations can be achieved.

[0101] The intelligent screening and evidence chain construction:

[0102] The Transformer model based on the multi-head attention mechanism realizes cross-document information association.

[0103] Automatically generate a problem evidence chain map, supporting tracing the problem back to the original document page number.

[0104] 3. A hierarchical data security protection module establishes a "five-level, three-domain" permission control model, dividing data access domains based on roles such as inspection teams, functional departments, and inspected units. Federated learning technology is used to enable cross-departmental data collaborative analysis, and problem clue data is encrypted and transmitted using zero-knowledge proofs. A data sandbox mechanism is implemented to ensure that problematic data is processed in a closed environment.

[0105] 4. Qualitative problem analysis module, which is used for qualitative problem analysis based on natural language processing and Transformer model-driven, including: multi-level semantic parsing, intelligent comparison of typical cases, and intelligent decision-making assistance.

[0106] The multi-level semantic analysis:

[0107] Use the BERT-CRF model for entity recognition and relationship extraction.

[0108] Transformer model based on hierarchical attention mechanism realizes long text encoding.

[0109] Intelligent comparison of typical cases:

[0110] Build a semantic retrieval library covering 3,000+ typical cases.

[0111] A multi-granularity semantic similarity calculation method is used to improve the comparison accuracy.

[0112] The intelligent auxiliary decision-making:

[0113] Automatically generate qualitative suggestion options for the problem based on the similarity calculation results.

[0114] Supports users to customize the comparison threshold and recommendation result sorting strategy.

[0115] 5. The inspection data analysis and intelligent report generation module establishes a three-level analysis system: "data-model-report," covering 18 core indicators, including petition reports, problem clues, and rectification effectiveness. It utilizes knowledge graph technology to construct a correlation network for inspection issues and automatically generates inspection analysis reports with five levels of risk warnings. It supports multi-dimensional data drilling, enabling interactive analysis by over 20 dimensions, including time, unit, and issue type.

[0116] The Transformer government inspection algorithm system is based on a customized training method for the Transformer model for government scenarios, a multi-granularity feature extraction algorithm for policy texts, a risk transmission path analysis model, and a dynamic context window optimization strategy.

[0117] The permission control is implemented based on the permission inheritance algorithm of the organizational structure, the temporary permission allocation and recovery mechanism, the data desensitization and encryption transmission solution, and the permission audit log analysis model.

[0118] System performance optimization is achieved based on distributed task scheduling algorithm, hot and cold data separation storage strategy, abnormal traffic automatic flow limiting mechanism and intelligent disaster recovery solution.

[0119] An embodiment of the present invention further provides a device for implementing government inspection based on information management and natural language processing, comprising: at least one memory and at least one processor;

[0120] The at least one memory is configured to store a machine-readable program;

[0121] The at least one processor is used to call the machine-readable program to implement the government inspection method based on information management and natural language processing described in the above embodiment.

[0122] An embodiment of the present invention further provides a computer-readable medium having computer instructions stored thereon. When executed by a processor, the computer instructions implement the government inspection method based on information management and natural language processing described in the above embodiment. Specifically, a system or device equipped with a storage medium can be provided, on which software program code implementing the functions of any of the above embodiments is stored, and a computer (or CPU or MPU) of the system or device can be caused to read and execute the program code stored in the storage medium.

[0123] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute part of the present invention.

[0124] Examples of storage media for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, the program code can be downloaded from a server computer via a communication network.

[0125] In addition, it should be clear that the functions of any of the above embodiments can be achieved not only by executing the program code read by the computer, but also by enabling the operating system operating on the computer to complete part or all of the actual operations based on the instructions of the program code.

[0126] In addition, it can be understood that the program code read from the storage medium is written into the memory provided in the expansion board inserted into the computer or into the memory provided in the expansion unit connected to the computer, and then based on the instructions of the program code, the CPU installed on the expansion board or expansion unit is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above embodiments.

[0127] The present invention has been shown and described in detail above through the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above multiple embodiments, those skilled in the art can know that the code review methods in the above different embodiments can be combined to obtain more embodiments of the present invention, and these embodiments are also within the scope of protection of the present invention.

Claims

1. A government inspection method based on information management and natural language processing, characterized in that: The implementation of this method includes: A microservices architecture is used to build a government inspection information management platform, enabling inspection teams to access the system in real time through multiple terminals. Requested data is encrypted using a national secret algorithm and then transmitted to a secure domain server via the government extranet. This platform also implements a standardized data interface with government systems and establishes a dynamic mapping mechanism across the three dimensions of "data-business-process." An intelligent screening mechanism driven by natural language processing and the Transformer model, including: construction of a knowledge base of inspection observation points, multimodal document parsing, intelligent screening, and evidence chain construction; Hierarchical data security protection: Establish a permission control model and divide data access domains according to roles; Qualitative problem analysis based on natural language processing and Transformer model-driven, including: multi-level semantic analysis, intelligent comparison of typical cases, and intelligent decision-making support; Analyze inspection data and generate intelligent reports: establish a three-level analysis system of "data-model-report"; use knowledge graph technology to build an inspection problem association network, and automatically generate inspection analysis reports including graded risk warnings; support multi-dimensional data drilling function, and conduct interactive analysis according to multiple dimensions.

2. A government inspection method based on information management and natural language processing according to claim 1, characterized in that: The inspection observation point knowledge base is constructed as follows: Integrate a database of typical inspection cases from the past five years and use the improved Transformer-XL model for long text pre-training; Construct a semantic feature vector library containing 128 types of typical problems, annotated with risk level and domain label metadata.

3. A government inspection method based on information management and natural language processing according to claim 1 or 2, characterized in that: The multimodal document parsing, Adopting OCR+NLP dual-engine architecture, it performs layout analysis, table recognition and semantic parsing on scanned documents; Combined with the domain-adaptive Transformer model, accurate parsing of vertical field texts including government terminology, policies and regulations can be achieved.

4. The government inspection method based on information management and natural language processing according to claim 1 is characterized in that: The intelligent screening and evidence chain construction, The Transformer model based on the multi-head attention mechanism realizes cross-document information association; Automatically generate a problem evidence chain map, supporting tracing the problem back to the original document page number.

5. The government inspection method based on information management and natural language processing according to claim 1 is characterized in that: The hierarchical data security protection, Federated learning technology is used to achieve cross-departmental collaborative data analysis, and problem clue data is transmitted using zero-knowledge proof encryption; a data sandbox mechanism is implemented to ensure that problem data is processed in a closed environment.

6. The government inspection method based on information management and natural language processing according to claim 1 is characterized in that: The permission control model is a hybrid permission model based on RBAC+ABAC, which is implemented through a permission inheritance algorithm based on the organizational structure, and establishes a permission audit log analysis model based on temporary permission allocation and recovery.

7. The government inspection method based on information management and natural language processing according to claim 1 is characterized in that: The qualitative analysis of the problem driven by natural language processing and Transformer model, The multi-level semantic analysis: Use the BERT-CRF model for entity recognition and relationship extraction; The Transformer model based on the hierarchical attention mechanism realizes long text encoding; Intelligent comparison of typical cases: Build a semantic retrieval library covering 3,000+ typical cases; Adopt multi-granularity semantic similarity calculation method to improve comparison accuracy; The intelligent auxiliary decision-making: Automatically generate qualitative suggestion options for the problem based on similarity calculation results; Supports users to customize the comparison threshold and recommendation result sorting strategy.

8. A government inspection system based on information management and natural language processing, characterized in that: include: Government Inspection Information Management Platform: This platform is built using a microservices architecture, supporting inspection teams to access the system in real time through multiple terminals. Request data is encrypted using a national encryption algorithm and then transmitted to a secure domain server via the government extranet. Implement standardized data interfaces with government systems and establish a dynamic mapping mechanism across three dimensions: data, business, and processes. An intelligent screening module based on natural language processing and Transformer model-driven, including: construction of a knowledge base of inspection observation points, multimodal document parsing, intelligent screening and evidence chain construction; Hierarchical data security protection module, used to establish permission control models and divide data access domains according to roles; Qualitative problem analysis module, which is driven by natural language processing and the Transformer model and includes multi-level semantic parsing, intelligent comparison of typical cases, and intelligent decision-making support. The inspection data analysis and intelligent report generation module establishes a three-level analysis system of "data-model-report", uses knowledge graph technology to construct an inspection problem association network, and automatically generates an inspection analysis report including graded risk warnings; The system implements government inspection work through the method described in any one of claims 1 to 7.

9. A device for implementing government inspection based on information management and natural language processing, characterized in that: include: at least one memory and at least one processor; The at least one memory is configured to store a machine-readable program; The at least one processor is configured to call the machine-readable program to implement the method according to any one of claims 1 to 7.

10. A computer-readable medium, characterized in that The computer readable medium stores computer instructions, which, when executed by a processor, can implement the method according to any one of claims 1 to 7.

Images