Automatic government affair approval system and method based on artificial intelligence

Abstract

The invention provides an automatic government affair examination and approval system and method based on artificial intelligence, and is used for disassembling a government affair examination and approval process into a plurality of function modules which can operate independently. Intelligent modeling is carried out on key links such as material processing, rule judgment, process planning and result generation related to government affair handling through artificial intelligence and data driving technologies, and an automatic, high-efficiency and interpretable whole examination and approval process is realized through cooperation among modules; therefore, a comprehensive solution integrating artificial intelligence information processing capability, a policy knowledge rule engine and a process automation mechanism is provided, the comprehensive advantages of flexible adaptation, efficient processing, traceable result, controllable strategy and the like are achieved, and the government affair approval system can be promoted to develop to a new efficient, fair and intelligent stage.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to an automatic government affair approval system and method based on artificial intelligence. BACKGROUND

[0002] Nowadays, the traditional approval mode relying on paper materials and manual circulation is gradually evolving into an electronic system. With the deployment of e-government platforms, through process modeling and information system integration, electronic operations such as business acceptance, data uploading, approval circulation, and result feedback are realized. Some matters have realized cross-departmental collaboration and parallel approval, which has preliminarily alleviated the problems of slow manual process and repeated submission of materials.

[0003] However, the present inventors have found that the existing government affair approval systems are mostly based on form driving and workflow engine, supporting business logic through configuration of approval nodes and data verification rules, usually providing standardized electronic forms for users to fill in application information, while supporting the uploading of attachment materials. After business acceptance, the staff performs approval judgment in the background and processes the approval results according to policies and regulations. Although such systems have improved the overall business operation efficiency, there are still many technical bottlenecks and operation limitations in the processing process, which are specifically limited in the following aspects: 1) Insufficient information extraction and structuring capability. The materials submitted by users are diverse, including pdf documents, scanned copies, handwritten application forms, and multi-format electronic spreadsheets. The system lacks the ability to recognize and structure unstructured data, resulting in manual review of materials and entry of key fields by approval personnel, increasing labor costs and error probability.

[0004] 2) Approval rule execution relies on manual judgment. Although some systems support rule configuration, most current rules are simple field verification or process branching judgment, which are difficult to support complex conditions, multi-level logic, and dynamic changes in policy scenarios. Approval personnel often rely on experience to understand regulations and make case-by-case judgments, resulting in inconsistent approval standards and inconsistent results.

[0005] 3) Limited cross-system data linkage capability. In matters handled by multiple departments, there are problems such as non-uniform data interface standards, information silos, and permission barriers between different systems. Approval personnel often need to manually retrieve external system data or verify applicant information, enterprise qualifications, historical records, and other content through offline methods, affecting the overall approval timeliness and accuracy.

[0006] 4) Lack of intelligent interaction and process guidance. Most current government service systems are based on static pages and fixed field inputs. Users have difficulty obtaining real-time feedback or personalized guidance during the process, and for applicants who are not familiar with the business process, it is easy to have problems such as misfilling and missing reports, which increases the number of communications and corrections and reduces the experience of doing business.

[0007] In addition, existing systems generally fail to integrate artificial intelligence (AI) technology capabilities. The deep semantic analysis of government data, intelligent identification of material content, automatic matching of policy provisions, and dynamic understanding of user intent are still weak, which makes it difficult for the system to achieve true automation and intelligent auxiliary judgment of approval, and it is also difficult to adapt to high-frequency handling, high-complexity judgment, and diversified user needs of government scenarios.

[0008] In summary, although the existing government approval technology system has achieved a certain degree of informatization and process automation, it still faces key problems such as insufficient intelligence, low system coordination efficiency, inconsistent rule execution, and poor user operation experience. A comprehensive solution that integrates artificial intelligence information processing capabilities, policy knowledge rule engines, and process automation mechanisms is needed to promote the development of government approval systems to a new stage of high efficiency, fairness, and intelligence. SUMMARY

[0009] The present application provides an automatic government approval system and method based on artificial intelligence, which is used to decompose the government approval process into multiple independently running functional modules, intelligently model the key links involved in government handling such as material processing, rule judgment, process planning, and result generation through artificial intelligence and data-driven technology, and realize the whole process of automatic, efficient, and interpretable approval through collaboration between modules. This provides a comprehensive solution that integrates artificial intelligence information processing capabilities, policy knowledge rule engines, and process automation mechanisms, with comprehensive advantages such as flexible adaptation, efficient processing, traceable results, and controllable strategies, which can promote the development of government approval systems to a new stage of high efficiency, fairness, and intelligence.

[0010] In a first aspect, the present application provides an automatic government approval system based on artificial intelligence. The automatic government approval system includes a data access module, an intelligent analysis module, a rule reasoning module, an approval path planning module, a process scheduling module, and a result generation and feedback module. Each module runs independently and collaborates through a standard interface and an event-driven communication mechanism. The data access module is used to receive the approval materials submitted by the user through the government service terminal in the preset material form and the preset material channel, and convert them into an internal standard data format. The intelligent analysis module is configured to perform text recognition on the approval materials, and perform approval-required field recognition on the basis of the text recognition result to obtain an approval-required field recognition result. The rule reasoning module is configured to perform compliance verification on the approval-required field recognition result. The approval path planning module is configured to perform approval path planning on the approval-required field recognition result that meets the compliance requirement to obtain a suitable approval path graph, the approval path graph being configured in a directed graph structure, a node representing an approval link or a responsible subject, and an edge representing an accessible path or a conditional branch. The process scheduling module is configured to drive execution of each node of the approval path on the basis of the approval path graph. The result generation and feedback module is configured to arrange an approval conclusion on the basis of the execution of the approval path graph, and perform multi-terminal push feedback.

[0011] In a second aspect, the application provides an automatic government affair approval method based on artificial intelligence, the method being applied to an automatic government affair approval system based on artificial intelligence, the automatic government affair approval system comprising a data access module, an intelligent analysis module, a rule reasoning module, an approval path planning module, a process scheduling module and a result generation and feedback module, each module being independently operated and cooperating through a standard interface and an event-driven communication mechanism, and the method comprising the following steps: The data access module receives approval materials submitted by a user through a government affair service terminal in a preset material form and a preset material channel, and converts the approval materials into an internal standard data format. The intelligent analysis module performs text recognition on the approval materials, and performs approval-required field recognition on the basis of the text recognition result to obtain an approval-required field recognition result. The rule reasoning module performs compliance verification on the approval-required field recognition result. The approval path planning module performs approval path planning on the approval-required field recognition result that meets the compliance requirement to obtain a suitable approval path graph, the approval path graph being configured in a directed graph structure, a node representing an approval link or a responsible subject, and an edge representing an accessible path or a conditional branch. The process scheduling module drives execution of each node of the approval path on the basis of the approval path graph. The result generation and feedback module arranges an approval conclusion on the basis of the execution of the approval path graph, and performs multi-terminal push feedback.

[0012] In a third aspect, the application provides a computer readable storage medium, the computer readable storage medium storing a plurality of instructions, the instructions being adapted to be loaded by a processor to execute the method provided in the second aspect of the application.

[0013] From the above content, the application has the following beneficial effects: The application disassembles the government examination and approval process into multiple independently operable function modules, intelligently models key links such as material processing, rule judgment, process planning and result generation involved in government handling through artificial intelligence and data driving technology, and realizes the automatic, high-efficiency and explainable whole process of examination and approval through the cooperation between modules, thus providing a comprehensive solution integrating the information processing ability of artificial intelligence, policy knowledge rule engine and process automation mechanism, having comprehensive advantages such as adaptive flexibility, efficient processing, traceable result and controllable strategy, and can promote the development of the government examination and approval system to a new stage of high efficiency, fairness and intelligence. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0015] Figure 1 A system architecture diagram of the automatic government examination and approval system based on artificial intelligence of the present application; Figure 2 A flowchart of the automatic government examination and approval method based on artificial intelligence of the present application. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0017] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. The naming or numbering of steps in this application does not mean that the steps in the method flow must be executed in the time / logical sequence indicated by the naming or numbering. The process steps that have been named or numbered can be changed in the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0018] The division of modules in this application is a logical division. In actual application, there may be other division methods. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection between modules can be electrical or other similar forms, which are not limited in this application. Moreover, the modules or submodules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed into multiple circuit modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application.

[0019] First, refer to Figure 1 The system architecture diagram of the artificial intelligence-based automated government approval system of the present application is shown. The artificial intelligence-based automated government approval system provided by the present application can specifically include several major modules: data access module, intelligent analysis module, rule reasoning module, approval path planning module, process scheduling module and result generation feedback module. Each module runs independently and collaborates with the event-driven communication mechanism through a standard interface.

[0020] At this point, it can be understood that this application breaks down the government approval process into multiple independently run functional modules, and uses artificial intelligence and data-driven technologies to intelligently model key links such as material processing, rule judgment, process planning and result generation involved in government affairs. Through collaboration between modules, the entire approval process is automated, efficient and explainable.

[0021] Among them, each module can realize event-driven communication through an asynchronous message bus. The event bus adopts a lightweight message queue protocol, such as Message Queuing Telemetry Transport (MQTT) or Kafka. Each module triggers corresponding computing logic after receiving a specific event and publishes the next stage event after processing is completed, thereby forming a loosely coupled process link. This mechanism can further improve the concurrency capability of the system and avoid overall process lagging caused by blocking of the approval node.

[0022] Each module is deployed in a distributed architecture and forms a clear boundary through modular service division. In the actual engineering landing process, a front-end and back-end separation architecture design is recommended. The front-end interface interacts with the back-end approval engine through HTTP or gRPC.

[0023] In this way, the overall architecture adopts a modular and distributed deployment mode. Each functional unit of the system runs independently and cooperates with each other through a standard interface and an event-driven communication mechanism to form a flexible and scalable approval service network. Each module automatically selects a subsequent processing path according to a configured service orchestration strategy or model recommendation after receiving the processing result from the previous module, thereby realizing the dynamic and nonlinear development of the approval process.

[0024] At the same time, in terms of specific implementation, the back-end part of the system mainly consists of several microservices, including material identification services, field extraction services, rule engine services, path planning services, scheduling control services, result generation services, and log archiving services. Each service is deployed as a containerized unit and runs on a cloud server or a local government affairs dedicated server node. Automatic scaling and high-availability deployment are realized through a service orchestration platform (such as a Kubernetes management platform). Each module communicates through RESTful API or message queue and has good horizontal expansion capability.

[0025] The storage layer of the system includes a relational database (such as PostgreSQL) for storing user materials, field structures, and approval process information. An object storage system is used to save uploaded image materials and intermediate results of identification. A log system (such as ELK or ClickHouse) is used to collect and analyze model behavior and operation traces.

[0026] The hardware platform on which each functional module runs includes but is not limited to: a multi-core central processing unit (CPU), a graphics accelerator (graphics processing unit (GPU) or neural processing unit (NPU)) for running related models, a storage unit for program and data storage, and a network interface for external system docking. The system supports running in a private deployment environment while ensuring functional accuracy, meeting the high requirements for information security and data isolation in government scenarios.

[0027] Next, the cooperation relationship between the modules and the modules will be described in detail from the specific module working process.

[0028] (1) Data access module The data access module is used to receive the approval materials submitted by the user through the government service terminal under the preset material form and the preset material channel, and convert them into internal standard data format.

[0029] It can be understood that the data access module corresponds to the input part of the system of the present application, which can allow the user to initiate the corresponding approval task and submit the corresponding approval materials through the relevant user interface. Of course, in actual situations, there are also cases where the system initiates the approval task or other people initiate the approval task, that is, the current approval task is not necessarily initiated by the user who submits the approval materials, which corresponds to the diversified application requirements in actual situations.

[0030] As an example, the user can submit the approval materials through the corresponding government service terminal.

[0031] In addition, it can be seen that the present application also determines different material forms and different material channels in detail in advance / previously, forms preset material forms and preset material channels, so as to meet the diversified and standardized approval material submission requirements.

[0032] As an exemplary embodiment here, for the data access module, the preset material form can specifically involve structured electronic forms, scanned copies and pdf documents, and the preset material channel can specifically involve web forms, WeChat mini programs and government apps, and after encryption and desensitization processing of sensitive fields including ID number and enterprise tax number, the original materials submitted by the user are converted into internal standard data format (which can be denoted as D_input) and tracked in the audit process through the assigned unique identification (Identification, ID).

[0033] It can be understood that encryption and desensitization by means of hashing or salting can further improve the information security of the system, and better protect user privacy. Under this setting, subsequent data processing or approval processing is easily understood as being carried out in an encrypted and desensitized state.

[0034] For data after desensitization, federated learning technology can be used to complete model collaborative training in subsequent model training steps without exposing raw data. This mechanism ensures data security while achieving cross-regional knowledge sharing.

[0035] In addition, the data access module can also preprocess the accessed data to enhance data quality, filter out abnormal data, or adapt to subsequent data processing.

[0036] For example, images such as business licenses, ID scans, and tax registration certificates can be preprocessed with denoising, correction, edge enhancement, and other specific operations.

[0037] For structured electronic forms, the field content can also be cleaned and standardized to generate a unified "field-value" mapping dictionary structure.

[0038] Of course, preprocessing can be handled by the data access module, or by the intelligent parsing module or other modules that follow. This is possible in actual situations and can be flexibly configured as needed.

[0039] (2) Intelligent parsing module The intelligent parsing module is used for text recognition of the approval materials, and performs approval-required field recognition based on the text recognition result to obtain an approval-required field recognition result.

[0040] It can be understood that the intelligent parsing module is used to process the approval materials accessed by the data access module into specific fields that can be processed by the subsequent rule reasoning module, i.e., the approval-required fields. This process realizes the conversion of raw materials to business semantic structures, laying a good foundation for intelligent approval.

[0041] This involves two stages of processing, the first stage being the text recognition stage, and the second stage being the structured extraction stage.

[0042] Specifically, in the first stage, for non-structured materials such as images and pdf documents, Optical Character Recognition (OCR) technology can be used to recognize the text involved.

[0043] As an example, the text recognition result can be configured as an output format of a two-dimensional character matrix M_text.

[0044] In the second stage, a deep neural network model based on multi-channel feature fusion can be used to realize accurate positioning and semantic classification of the field by a pre-trained named entity recognition model and a pattern matching mechanism, combined with multi-dimensional features such as morphology, syntax, and entity context, to extract subsequent required fields for approval, such as specific fields of subject name (such as enterprise name), subject type, matter category, certificate type, handling area, validity period, application time, contact information, unified social credit code, registered capital, legal representative, etc.

[0045] The form of the extraction function can be expressed as follows: , Where F represents a structured field set, which is the core input for subsequent rule reasoning and process planning, represents the input text, represents the extraction function.

[0046] In addition, the intelligent analysis module here can also support noise reduction and word segmentation for redundant fields, and disambiguation processing for ambiguous expressions in long texts, so as to ensure that the field extraction accuracy meets the minimum information unit definition required for approval.

[0047] All extracted fields are uniformly converted into an approval field set as the input data structure for subsequent processing. Each field in the field set is attached with field type, extraction source, confidence, and whether it is manually confirmed, etc. properties. The system can set a prompt marker for low-confidence fields for manual correction or trigger a supplementary task.

[0048] (3) Rule reasoning module The rule reasoning module is used to check the compliance of the identified fields required for approval.

[0049] It can be understood that the compliance check here is to check whether the content of the materials meets the requirements, for example: if the applicant is a natural person subject, the system judges whether to provide valid identification; if the application matter is qualification change, the system checks whether the original certificate and change basis are uploaded.

[0050] All aspects of the check results or judgment results can be summarized into a compliance status list, including: whether it meets the approval requirements, whether there is a missing field, a list of missing fields, the type and description of materials to be supplemented, and subsequent processing suggestions.

[0051] It can be understood that, for non-compliant places, by making a prompt in a visual and highly visible manner, real-time feedback can be formed to allow the user to make corresponding processing to pass the compliance check.

[0052] In terms of details, the legal judgment rules required by the relevant approval business that the rule inference module needs to follow in the specific compliance check process can be stored in a rule base, which can be stored in a domain-specific language (DSL) manner, supporting combined logic expression. Rule matching is matched with field values through the construction of a Boolean expression, and an output judgment flag J is outputted, which is used to indicate whether the field is compliant, missing, or in conflict, etc. The module has the ability of dynamic rule loading and version management, ensuring that different policies in different places can be flexibly adapted. Its judgment logic is expressed as: , Where J is the inference result vector, R represents the rule set of the current business type, supporting structured mark output, and F represents the input structured field set.

[0053] Further, as an exemplary embodiment herein, for the rule inference module, each rule is composed of logical expressions, weight conditions, mandatory check, cross-field logic verification, and dependency relationships, etc. elements bound by field comparison for compliance judgment, and can specifically involve qualification verification rules, regional practicality judgment rules, applicant authority determination rules, and material completeness judgment rules. The rule expression adopts the following structure: , Where R represents the rule set; represents the i-th rule; represents the total number of rules; represents the structured field; represents the rule function judgment condition, when the result is 1, it means that the rule is met, and when the result is 0, it means that it is not met.

[0054] (4) Approval path planning module The approval path planning module is used to identify the results of the fields required for the approval of the compliance requirements, and to obtain a suitable approval path graph. The approval path graph is configured in a directed graph structure, where the nodes represent the approval links or responsible subjects, and the edges represent the reachable paths or conditional branches (or sequential dependencies or parallel relationships).

[0055] It can be understood that the approval path planning module here is to build the approval process involved in the current approval task in a popular way, and it specifically involves a directed graph in a specific form, that is, by integrating historical data, business configuration and AI model output, the optimal path is inferred by using graph search and strategy prediction model.

[0056] Specifically, the approval path planning module can automatically generate a corresponding approval execution path based on the application matter type, compliance status and business library preset process model. In detail, the path is composed of a group of nodes, each node represents an approval subtask, and the node types include automatic approval nodes, manual review nodes, mixed processing nodes and the like. The system configures trigger conditions, processing content and output structure for each node. The approval path supports serial and parallel mixed structure, and the path process is represented by a graph structure. The dependency relationship between each node can be defined by a logical expression, such as: "if the materials are complete, skip the manual verification node and directly enter the system verification node".

[0057] The approval path planning module can also statistically evaluate whether the approval path is optimal by counting the number of nodes, node type combination and approval time limit of each generated approval path graph, and continuously update the path generation strategy through reinforcement learning algorithm, so as to select a more efficient path graph structure in the next round of similar matter approval.

[0058] In addition, path visualization and decision explanation can also be supported here to facilitate supervision of the approval process.

[0059] The path function is represented as follows: , Where G represents the generated approval path graph, F represents the input structured field set, J represents the inference result vector, and F represents the path function.

[0060] Specifically, as an exemplary embodiment, for the path planning strategy model, it specifically constructs a state space based on a graph neural network, and determines a suitable path structure by combining path cost and strategy preference; The path cost function can be represented by the following formula: , Where, represents the total cost of the path P; represents the average processing time of the i th link; represents the link weight; represents the total number of links; represents the path stability adjustment coefficient; represents the total influence value of uncertain factors in the path.

[0061] It can be understood that the path cost as a reference factor involved in the approval path planning based on the directed graph is given a set of practical implementation schemes combined with specific quantitative formulas.

[0062] (5) Process scheduling module The process scheduling module is used to drive the execution of each node of the approval path based on the approval path graph.

[0063] It can be understood that the process scheduling module is used to promote the approval processing of different approval nodes planned. The scheduling logic supports an event-driven mechanism, that is, when a node completes processing, the system automatically publishes the processing result to the downstream node to trigger the next approval. The processing mode of the node includes manual confirmation, AI approval suggestion, automatic decision, etc. The process state is managed by a finite state machine model, and the transfer function can be represented as: , Where, represents the current state at time t, represents the trigger event at time t, represents the transfer function, and the system dynamically updates the state based on the policy rule set.

[0064] In this way, the approval path graph constructed in the previous step is sent to the process scheduling module to start processing the approval task node by node. Whenever an approval node is activated, the system determines the node type and selects the corresponding execution logic.

[0065] For an automated approval node, the system calls the program to perform field verification, material consistency comparison, external interface verification, and other logical tasks, and records the processing results. For a manual approval node, the system pushes the task to the government background personnel through the interface for processing. The manual approval node returns the results to the system through the interface after processing is completed.

[0066] The execution results, processing time, status code, and exception information of each approval node are recorded to ensure the traceability and closed-loop control of the path process.

[0067] (5) Result generation and feedback module The result generation and feedback module is used to summarize the approval conclusion based on the execution of the approval path graph and to push feedback to multiple ends.

[0068] It is easy to understand that the result generation feedback module is responsible for the final processing link of the approval task, and arranges the approval conclusion based on the execution state, output data and final summary judgment of each approval node. In this process, it can involve specific approval conclusion information such as approval status (pass, not pass, return for modification, etc.), approval basis chain (including key judgment link), rule set, approval time record, material integrity explanation (suggested supplementary materials), etc. The final result is composed of structured data and natural language explanation, and is pushed to the user and the business system to support the generation of legally effective approval documents through electronic signature.

[0069] All approval results will be pushed to the business system database through the synchronization interface for registration and archiving, and users can also obtain the approval status, progress node and feedback content in real time through the approval query interface. At the same time, according to the actual needs of government affairs services, the output results can be output to the SMS service, government app push channel or SMS reminder interface to realize the multi-end notification function.

[0070] In addition, as an example, for the result generation feedback module, in addition to outputting the approval result, it can also generate auxiliary explanation text to help users understand the logical basis of the approval conclusion.

[0071] For example, for the case of disapproved approval, the system will automatically list the denied rule items, corresponding fields and suggested supplementary material explanations in combination with the rule matching path.

[0072] This explanation text is generated based on the business language expression template on the basis of the large language model completion method, which can be understood as a processing explanation. Without involving the core algorithm details, it is presented in the "because…so…" structure, so as to effectively improve the approval transparency and user satisfaction.

[0073] Further, as an exemplary embodiment, the automatic government affair approval system based on artificial intelligence of the present application can also include a training management module; The training management module is used for unified management of the text recognition model for text recognition of the intelligent analysis module, the named entity recognition model for approval required field recognition of the intelligent analysis module, the rule adaptation model for compliance verification of the rule reasoning module, and the path planning strategy model for approval path planning of the approval path planning module. The management includes training, evaluation and updating respectively to maintain the adaptability of the system in different government affairs scenarios.

[0074] In specific operation, the training management module can also construct training samples by collecting historical approval data, user interaction behavior, rule adaptation feedback and other information, and continuously optimize model performance through supervised learning, transfer learning and reinforcement learning methods.

[0075] Further, for the named entity recognition model involved, the following loss function can be used in the training process: , wherein, represents the total loss of the model, is used to measure the difference between the predicted result and the true label; n represents the length of the input sequence, represented by the total number of words or words in the text; C represents the number of categories, including names, places, organizations and others; represents whether the true label of the i-th word or word belongs to the c-th category, if it belongs to, it is 1, otherwise it is 0; represents the probability that the model predicts that the i-th word or word belongs to the c-th category, and is obtained by the softmax operation.

[0076] It can be understood that the present application sets up here for the loss function involved in the model training link, and gives a set of highly adaptive and practical application scheme for the named entity recognition model.

[0077] Similarly, in another aspect, for the path planning strategy model, the following loss function can be used in the training process: , wherein, represents the path expectation deviation; represents the actual path sequence vector; represents the model predicted path; N represents the total number of training samples, which can effectively improve the prediction accuracy and execution efficiency of the approval path.

[0078] In addition, in addition to the training management module, as an exemplary embodiment, the artificial intelligence-based automatic government approval system of the present application can also include a log audit module.

[0079] The log audit module is used for full-process audit log recording of each approval request, that is, full-process recording of each approval request in the form of audit log, which can involve data access time, processing steps, rule hit situation, approval person suggestion, system suggestion, final result, original material content, field extraction result, rule judgment path, approval node order, node result state, artificial intervention behavior record and user feedback content, etc., and is stored in a tamper-proof chain structure, supporting supervision platform reading, behavior replay and traceability.

[0080] The system can periodically trigger the model fine-tuning training process based on the above log data. Taking the named entity recognition model as an example, the system can compare the differences between the model extraction field and the final approval personnel modified field to generate training sample pairs for fine-tuning the model; the subsequent rule matching part, i.e. the rule adaptation model, can also adjust and optimize the rule expression according to the manually supplemented field of the approval personnel and the rule feedback.

[0081] Among them, the log audit module can also periodically interact with the previous training management module to use the historical processing behavior as a kind of training sample for the related model training work of the training management module, so as to realize the closed-loop optimization of system capability.

[0082] At the same time, in the above-mentioned system working process or approval flow process, the system can also judge whether there is abnormal data or rule conflict, if there is uncertain situation or high-risk approval behavior, the system will trigger the manual review mechanism, which can be easily understood. The current approval context information is pushed to the manual operation platform, and the subsequent automatic processing process is suspended until the manual operation is clear. While ensuring the efficient operation of the system, the human-computer cooperation capability is retained to deal with complex situations, which can be considered as a bottom-up measure.

[0083] At the same time, the system can also have self-learning ability, the core of which is the approval feedback learning module of the training management module. Through the result generation feedback module in further feedback collection and processing, the approval result backend data, user complaints, manual review opinions and other feedback collection information can be recorded and converted into a kind of training sample, which is provided for the related model training work of the training management module. In this way, the system can also periodically retrain the model and fine-tune the parameters to improve the accuracy, rationality and user satisfaction of the approval.

[0084] In this way, taking the closed-loop learning mechanism here as an example, the entire system of the present application has high data closed-loop capability and can realize the continuous iterative closed loop of "user material submission→intelligent judgment processing→result output→result feedback→data archiving learning→model optimization update".

[0085] Finally, for the above system settings, in general, the application will break down the government approval process into multiple independently running functional modules, intelligently model the key links such as material processing, rule judgment, process planning and result generation involved in government handling through artificial intelligence and data-driven technology, and realize the automatic, efficient and interpretable whole process of approval through the cooperation between modules, thus providing a comprehensive solution that integrates the information processing capability of artificial intelligence, policy knowledge rule engine and process automation mechanism, with comprehensive advantages such as flexible adaptation, efficient processing, traceable results and controllable strategy, which can promote the development of government approval system to a new stage of high efficiency, fairness and intelligence.

[0086] And in detail, there are: 1) By introducing artificial intelligence technology, the automation level of government approval process is significantly improved.

[0087] In the traditional government approval process, material identification, information comparison, rule judgment and process transfer mostly rely on manual operation, which not only has long processing period and high error rate, but also is difficult to adapt to the scene of high concurrency and multiple types of business applications.

[0088] In contrast, by integrating image recognition, natural language processing, rule reasoning and path optimization algorithms, the application can automatically complete the complete process from material receiving to approval conclusion output without human intervention, greatly improving the approval efficiency and reducing the labor cost.

[0089] 2) The modular architecture proposed in the application has good system scalability and flexibility.

[0090] Each module works together through standard interfaces and asynchronous communication mechanisms, so that the system can quickly adapt or replace part of the logic components when dealing with different government business scenarios without the need for overall reconstruction.

[0091] At the same time, the architecture supports micro-service deployment, which is convenient for later system maintenance, function iteration and cross-region deployment, and meets the needs of local policy differences while realizing unified platform support.

[0092] 3) Through the cooperation of rule engine and approval path planning model, dynamic adaptation of complex approval rules and individual customization of approval process are realized.

[0093] When facing different types of approval matters, the system can automatically judge the applicable rules based on the material content submitted by the user, combine the approval nodes, and predict the optimal approval path, so as to ensure that the process is legal and efficient, controllable, which can effectively reduce the problems of repeated approval, backtracking and other problems caused by inconsistent understanding of rules or unreasonable node setting, and improve the transparency and continuity of the approval process.

[0094] 4) The feedback learning mechanism introduced in this application builds an adaptive closed-loop optimization system.

[0095] Through continuous collection of historical approval data, user interaction records, and artificial review opinions, the system can periodically optimize related models, continuously improving accuracy and adaptability in actual operation.

[0096] This feature enables the system to have a self-evolution ability of "the more you use, the more accurate it becomes", which is particularly suitable for practical application environments where government rules change frequently and business scenarios are highly diversified.

[0097] It also strengthens the compliance of data processing and the explainability of results. In the approval process, the system generates processing records and explanations for each decision node, facilitating supervision and user complaints.

[0098] At the same time, the log audit module provides a full-process traceable approval track, combined with the model explainable algorithm, which can realize manual confirmation and responsibility tracing of intelligent decision results, solving the risk concerns caused by the "black box" of current intelligent systems.

[0099] 4) It has obvious advantages in improving the standardization and consistency of government services.

[0100] By modeling the approval rules in a structured and executable logical expression, the system can maintain consistent rule application standards among different service windows and different operators, effectively eliminating the inconsistency of results caused by understanding bias and operational differences in traditional manual approval.

[0101] This standardization capability not only enhances the standardization of government services, but also provides important technical support for cross-regional government collaboration and "one-stop service" policies.

[0102] At the same time, it realizes the fusion and processing capability of multi-modal government data, which has a significant advantage in dealing with complex application materials.

[0103] Traditional government systems rely on structured forms, while real-world government materials often exist in the form of scans, paper proofs, and unstructured text. By integrating text recognition, natural language analysis, entity recognition, and semantic analysis algorithms, it can deeply analyze and extract key elements from mixed layout, multi-format input materials, and build a unified data representation format, thereby bridging the "last mile" of government data and significantly improving the system's intelligent understanding ability and material processing breadth.

[0104] 6) This application also has a positive effect on improving user experience.

[0105] By constructing the front-end intelligent guidance mechanism and the back-end personalized recommendation mechanism, the system can prompt the missing fields, provide demonstration content, and predict possible non-compliance items during the user's submission of materials or filling out of forms, thereby reducing the difficulty of filling out and the return due to incomplete information.

[0106] At the same time, the approval feedback information is not limited to the conclusion of pass or fail, but also includes the approval basis, rule hit explanation and supplementary suggestions, so that the user can clearly understand the handling process and improve the trust and satisfaction of government service.

[0107] 7) The present application supports multi-channel and multi-terminal access to government approval processes, with good service coverage.

[0108] The system can be deployed on various government service terminals, including government websites, government apps, WeChat public accounts, mini programs, offline self-service terminals, etc., to realize online and offline integrated access. In different terminal environments, the system ensures consistent approval processes and synchronous data processing through unified data specifications and service interfaces, which helps to promote the comprehensive digital transformation and intelligent upgrade of government services.

[0109] 8) The module decoupling feature of the present application also makes it have good engineering maintainability and technical sustainability.

[0110] In actual deployment, the system can flexibly adjust the operation logic of each module according to business load, actual demand in the region, or policy evolution, which not only facilitates fault location and performance tuning, but also reduces the development and testing cost when new policies are online, avoiding the problems of update lag and complex adaptation commonly seen in traditional centralized systems.

[0111] The above is the introduction of the automatic government approval system based on artificial intelligence provided by the present application, and on the basis of the system, the present application also provides an automatic government approval method based on artificial intelligence. Obviously, the method is applied to the automatic government approval system based on artificial intelligence, and the automatic government approval system based on artificial intelligence briefly includes data access module, intelligent analysis module, rule reasoning module, approval path planning module, process scheduling module and result generation feedback module. Each module runs independently and cooperates through standard interface and event-driven communication mechanism.

[0112] Based on the above system design, the automatic government approval method based on artificial intelligence provided by the present application refers to Figure 2 The present application shows a flowchart of the automatic government approval method based on artificial intelligence, which can specifically include the following steps S201 to S206: Step S201, the data access module receives the approval materials submitted by the user through the government service terminal under the preset material form and the preset material channel, and converts them into an internal standard data format; Step S202, the intelligent analysis module performs text recognition on the approval materials, and performs approval-required field recognition on the basis of the text recognition result to obtain an approval-required field recognition result; Step S203, the rule reasoning module performs compliance verification on the approval-required field recognition result; Step S204, the approval path planning module plans an approval path for the approval-required field recognition result that meets the compliance requirement, and obtains a suitable approval path graph. The approval path graph is configured in a directed graph structure, wherein a node represents an approval link or a responsible subject, and an edge represents a reachable path or a conditional branch; Step S205, the flow scheduling module drives the execution of each node of the approval path based on the approval path graph; Step S206, the result generation and feedback module summarizes the approval conclusion based on the execution of the approval path graph, and performs multi-end push feedback.

[0113] In an exemplary embodiment, for the data access module, the preset material form involves structured electronic forms, scanned copies and pdf documents, the preset material channel involves web forms, WeChat mini programs and government apps, and the original materials submitted by the user are converted into an internal standard data format after encryption and desensitization processing of sensitive fields including an ID number and a business tax number, and are tracked in the audit process by assigning a unique identifier.

[0114] In another exemplary embodiment, the automatic government approval system further comprises a training management module; The method further comprises: The training management module manages the text recognition model of the intelligent analysis module for text recognition, the named entity recognition model of the intelligent analysis module for approval-required field recognition, the rule adaptation model of the rule reasoning module for compliance verification, and the path planning strategy model of the approval path planning module for approval path planning, respectively, including training, evaluation and updating.

[0115] In another exemplary embodiment, for the named entity recognition model, the following loss function is used in the training process: , wherein, represents the total loss of the model, is used to measure the difference between the prediction result and the true label; n represents the length of the input sequence, which is represented by the total number of words or characters in the text; C represents the number of categories, including names, places, organizations and others; represents whether the true label of the i-th word or word belongs to the c-th class, and if it belongs to the c-th class, it is 1, otherwise it is 0; represents the probability that the model predicts that the i-th word or word belongs to the c-th class, and is obtained by the softmax operation.

[0116] In yet another exemplary embodiment, for the path planning strategy model, a state space is constructed based on a graph neural network, and a suitable path structure is determined by combining path cost and strategy preference; The path cost function is represented by the following formula: , wherein, represents the total cost of the path P; represents the average processing time of the i-th link; represents the link weight; represents the total number of links; represents the path stability adjustment coefficient; represents the total influence value of uncertain factors in the path.

[0117] In yet another exemplary embodiment, for the path planning strategy model, the following loss function is used in the training process, and the following path expectation deviation is used as the optimization target: , wherein, represents the path expectation deviation; represents the actual path sequence vector; represents the path predicted by the model; N represents the total number of training samples.

[0118] In yet another exemplary embodiment, for the rule reasoning module, each rule is composed of a logical expression, a weight condition, a mandatory check, a cross-field logical verification, and a dependency binding, compliance is judged by field comparison, and specifically involves qualification verification rules, regional practicality judgment rules, applicant permission determination rules, and material completeness judgment rules; The rule expression adopts the following structure: , wherein, represents a rule set; represents the i-th rule; represents the total number of rules; represents a structured field; represents a rule function judgment condition, when the result is 1, it means that the rule is satisfied, and when the result is 0, it means that it is not satisfied.

[0119] In yet another exemplary embodiment, the automatic government affair approval system further comprises a log auditing module; The method further includes: The log auditing module performs full-process auditing and logging for each approval request, involving data access time, processing steps, rule hit condition, approver suggestion, system suggestion, final result, original material content, field extraction result, rule judgment path, approval node order, node result state, manual intervention behavior record, and user feedback content, and stores the same in a tamper-proof chain structure, supporting reading, behavior replay, and traceability by a supervision platform.

[0120] Those skilled in the art can clearly understand the specific working process of the above-described artificial intelligence-based automated government approval method for the sake of convenience and brevity, which can be referred to as Figure 1 The description of the artificial intelligence-based automated government approval system in the corresponding embodiments will not be repeated here.

[0121] Those skilled in the art can understand that all or part of the steps in the various methods of the above-described embodiments can be completed by instructions or by relevant hardware controlled by the instructions, which can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0122] To this end, the present application provides a computer-readable storage medium, which stores a plurality of instructions capable of being loaded by a processor to execute the present application as Figure 2 The steps of the artificial intelligence-based automated government approval method in the corresponding embodiments can be specifically implemented by referring to the above description. Figure 2 The description of the artificial intelligence-based automated government approval method in the corresponding embodiments will not be repeated here.

[0123] The computer-readable storage medium can include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0124] Due to the instructions stored in the computer-readable storage medium, the present application as Figure 2 The steps of the artificial intelligence-based automated government approval method in the corresponding embodiments can be implemented by referring to the above description. Figure 2 The artificial intelligence-based automated government approval method in the corresponding embodiments can achieve the beneficial effects as described above, and the details are described above and will not be repeated here.

[0125] The above provides a detailed introduction to the automatic government affair examination and approval system based on artificial intelligence, the automatic government affair examination and approval method based on artificial intelligence and the computer readable storage medium provided by the application. The principle and implementation mode of the application are described by applying specific examples. The above description of the examples is only used to help understand the core idea of the application. Meanwhile, for those skilled in the art, the specific implementation mode and application range will be changed according to the idea of the application. In conclusion, the content of the specification should not be understood as a limitation of the application.

Claims

1. An automated government affairs approval system based on artificial intelligence, characterized in that: The automated government approval system includes a data access module, an intelligent analysis module, a rule reasoning module, an approval path planning module, a process scheduling module, and a result generation and feedback module. Each module operates independently and collaborates with an event-driven communication mechanism through a standard interface. The data access module is used to receive the approval materials submitted by the user through the government service terminal under the preset material form and preset material channel, and convert them into the internal standard data format; The intelligent parsing module is used to perform text recognition on the approval materials, and identify the fields required for approval based on the text recognition results to obtain the recognition results of the fields required for approval; The rule reasoning module is used to perform compliance verification on the identification results of the fields required for approval; The approval path planning module is used to plan the approval path based on the identification results of the approval required fields that meet the compliance requirements, and obtain a suitable approval path graph. The approval path graph is configured in a directed graph structure, where nodes represent approval links or responsible entities, and edges represent reachable paths or conditional branches; The process scheduling module is used to drive the execution of each node of the approval path based on the approval path diagram; The result generation feedback module is used to organize the approval conclusions based on the execution status of the approval path diagram and push feedback to multiple terminals.

2. The automated government affairs approval system according to claim 1, characterized in that: For the data access module, the preset material forms involve structured electronic forms, scans and PDF documents, and the preset material channels involve web forms, WeChat applets and government apps. The original materials submitted by users are converted into the internal standard data format after encryption and desensitization of sensitive fields including ID card number and corporate tax number, and are tracked during the audit process through the assigned unique identifier.

3. The automated government affairs approval system according to claim 1, characterized in that: The automated government affairs approval system also includes a training management module; The training management module is used to manage the text recognition model used by the intelligent analysis module for text recognition, the named entity recognition model used by the intelligent analysis module for identifying fields required for approval, the rule adaptation model used by the rule reasoning module for compliance verification, and the path planning strategy model used by the approval path planning module for approval path planning, including training, evaluation and updating.

4. The automated government affairs approval system according to claim 3, characterized in that: For the named entity recognition model, the following loss function is used during training: ， in, represents the total loss of the model, Used to measure the difference between the predicted result and the true label; n represents the length of the input sequence, expressed as the total number of characters or words in the text; C represents the number of categories, including names, places, organizations, and others; Indicates whether the true label of the i-th character or word belongs to the c-th category. If so, it is 1, otherwise 0; The representation model predicts the probability that the i-th character or word belongs to the c-th category, which is obtained by the softmax operation.

5. The automated government affairs approval system according to claim 3 is characterized in that: For the path planning strategy model, a state space is constructed based on a graph neural network, and a suitable path structure is determined by combining path cost with strategy preference. The path cost function is expressed as follows: ， in, represents the total cost of path P; represents the average processing time of the i-th link; represents the link weight; Indicates the total number of links; represents the path stability adjustment coefficient; Indicates the total impact value of uncertain factors in the path.

6. The automated government affairs approval system according to claim 5, characterized in that: For the path planning strategy model, the following loss function is used during training, and the following path expectation deviation is used as the optimization target: ， in, represents the expected deviation of the path; Represents the actual path sequence vector; Represents the model prediction path; N represents the total number of training samples.

7. The automated government affairs approval system according to claim 1, characterized in that: For the rule reasoning module, each rule consists of a logical expression, weight conditions, mandatory verification, cross-field logic verification, and dependency binding. Compliance is judged through field comparison, and specifically involves qualification verification rules, regional practicality judgment rules, applicant authority judgment rules, and material completeness judgment rules; A regular expression has the following structure: ， in, Represents a set of rules; represents the i-th rule; Indicates the total number of rules; Represents a structured field; Indicates the judgment condition of the rule function. When the result is 1, it means the rule is satisfied. When the result is 0, it means it is not satisfied.

8. The automated government affairs approval system according to claim 1, characterized in that: The automated government affairs approval system also includes a log audit module; The log audit module is used to record the entire process of audit logs for each approval request, including data access time, processing steps, rule hit status, approver suggestions, system suggestions, final results, original material content, field extraction results, rule judgment path, approval node sequence, node result status, manual intervention behavior records and user feedback content, and uses an unalterable chain structure for storage, supporting supervision platform reading, behavior replay and traceability.

9. An automated government affairs approval method based on artificial intelligence, characterized in that: The method is applied to an automated government approval system based on artificial intelligence. The automated government approval system includes a data access module, an intelligent analysis module, a rule reasoning module, an approval path planning module, a process scheduling module, and a result generation and feedback module. Each module operates independently and collaborates with an event-driven communication mechanism through a standard interface. The method includes: The data access module receives the approval materials submitted by users through the government service terminal in the preset material form and preset material channel, and converts them into the internal standard data format; The intelligent parsing module performs text recognition on the approval materials and, based on the text recognition results, identifies the fields required for approval to obtain the recognition results of the fields required for approval; The rule reasoning module performs compliance verification on the identification results of the fields required for approval; The approval path planning module plans the approval path based on the identification results of the approval fields that meet the compliance requirements, and obtains a suitable approval path diagram. The approval path diagram is configured as a directed graph structure, with nodes representing approval links or responsible entities, and edges representing reachable paths or conditional branches. The process scheduling module drives the execution of each node in the approval path based on the approval path diagram; The result generation feedback module organizes the approval conclusions based on the execution of the approval path diagram and pushes feedback to multiple terminals.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor to execute the method according to claim 9 .

Images