Civil aviation regulation and manual data structured governance method and system fusing ai and ocr

Abstract

The application relates to the technical field of civil aviation information management, in particular to a civil aviation regulation and manual data structured management method and system fusing AI and OCR. The method comprises the following steps: data acquisition and preprocessing; layout analysis and information positioning; semantic analysis and rule network construction; fracture detection and structured output. The application fuses AI semantic analysis and OCR text extraction technology, completes preprocessing, layout analysis and clause structured splitting on civil aviation regulation and manual documents, realizes deep semantic analysis of clauses based on a civil aviation domain ontology library, constructs entity, obligation modal and logical correlation, builds fine condition dependency relationship, dependency graph and computable rule chain network, detects rule chain logical consistency, identifies logical loss and dependency anomaly and intelligently completes or standardizes marks, generates a rule execution path, outputs structured data, and can also realize full-process iterative updating and data quality checking according to regulation revision.

Description

Technical Field

[0001] This invention relates to the field of civil aviation information management technology, and more specifically, to a structured governance method and system for civil aviation regulations and manual data that integrates AI and OCR. Background Technology

[0002] Civil aviation regulations, rules, and operation manuals are fundamental documents for ensuring safe aviation operations, and the industry generally relies on digital means to manage these documents. With the increasing maturity of OCR text recognition and artificial intelligence semantic analysis technologies, a feasible technical path has been provided for the intelligent analysis and associated management of unstructured aviation regulatory documents.

[0003] In existing technologies, research has been conducted in the fields of document structuring and regulatory graph construction. For example, invention patent CN202411156202.5 discloses a method and system for constructing a knowledge lineage graph for the aviation industry based on a large model, including the following steps: collecting policy and regulatory data; parsing and identifying relationships between documents; constructing and displaying the lineage graph; providing compliance review and feedback; and implementing system updates and maintenance. Another example is invention patent CN202110744524.1, which discloses a method and system for constructing a legal rule graph based on regulatory texts. It designs an information model for the legal rule graph, proposes a method for constructing legal terms, a classification method for identifying concepts, principles, and rules, and a method for identifying rule types. Ultimately, it realizes the construction of the legal rule graph and outputs an XML structured document, enabling computer systems to effectively utilize the legal rules existing in laws and regulations to carry out legislative, law enforcement, and judicial applications. It also plays a role in areas such as error correction of judgment documents, quality evaluation of judgment documents, case quality supervision, and legal provision dissemination in judicial case handling.

[0004] However, existing technologies still have the following technical shortcomings: Most graph construction solutions in the aviation and legal fields only achieve shallow structural modeling of overall document association or general legal provisions, lacking the ability to deeply depict the logical relationships of individual civil aviation regulations. This makes it difficult to support the industry's essential needs for clause logic compliance verification, assessment of the impact of regulatory revisions, and precise tracing of clause relationships. For the aviation policy knowledge lineage graph solution disclosed in CN202411156202.5, it only focuses on tracing the source and evolution of policy documents, without deeply exploring the inherent logical constraints and relationships between individual regulatory provisions. The modeling granularity is too coarse to meet the needs of refined clause logic management. For the legal rule graph construction solution disclosed in CN202110744524.1, its applicable scenario is the judicial application of general laws and regulations. It does not combine the specific business characteristics of the civil aviation industry for customized semantic modeling and relationship mining, and can only complete basic rule classification and standardized document output. It cannot adapt to the actual business scenarios of civil aviation regulation clause logic verification and the transmission of revision impact analysis. In light of this, we propose a structured governance method and system for civil aviation regulations and manuals data that integrates AI and OCR. Summary of the Invention

[0005] The purpose of this invention is to provide a structured governance method and system for civil aviation regulations and manuals data that integrates AI and OCR, in order to solve the problems of the graph construction solutions proposed in the background technology in the aviation and legal fields, which mostly only achieve shallow structured modeling of overall document association or general legal provisions, lack the ability to deeply depict the logical relationship of individual civil aviation regulations, and are unable to support the industry's urgent needs for logical compliance verification of provisions, assessment of the scope of impact of regulatory revisions, and accurate tracing of the relationship between provisions.

[0006] To address the aforementioned technical problems, one objective of this invention is to provide a structured governance method for civil aviation regulations and manual data that integrates AI and OCR, including: S1. Data Acquisition and Preprocessing: Preprocess the images of civil aviation regulations and manuals documents, and extract the text data from the documents using OCR technology; S2. Page Layout Analysis and Information Location: Based on text data, the document's page layout structure is identified and clause areas are divided to obtain structured clause units; S3. Semantic parsing and rule network construction: The structured clause unit is semantically parsed using an AI-based semantic parsing model to construct a semantic representation containing entity information, conditional relationships, and obligation modalities. The semantic representation is semantically constrained based on the civil aviation ontology library to construct conditional dependencies between clauses. Based on the conditional dependencies, a conditional dependency graph and the corresponding rule chain network are constructed to make the rule chain network computable and generate corresponding semantic evaluation information. S4. Break Detection and Structured Output: Based on the condition dependency graph, logical consistency detection is performed on the rule chain network to identify breaks in condition dependencies between clauses and incomplete rule execution paths; broken or abnormal rule chains are completed or marked based on the semantic evaluation results generated during the semantic parsing process; rule execution path information is generated based on the condition dependency graph; standardized structured data and rule execution path information are output, which are used to support the analysis of the propagation of the impact of clause revisions and the tracing of condition dependency paths; S5. Data Validation and Dynamic Update: Validate and assess the quality of structured data and rule execution path information, and trigger iterative updates of steps S1 to S4 based on revisions to civil aviation regulations and manuals.

[0007] As a further improvement to this technical solution, the layout analysis and information positioning in S2 includes the following steps: S2.1 Based on text data, layout structure recognition technology is used to identify structural elements in the document and generate layout metadata containing position coordinates and hierarchical relationships; among which structural elements include heading levels, chapter numbers, clause numbers, paragraph boundaries, table structures, and appendix markers, etc. S2.2. Use page layout metadata to divide the document into clause areas, cut continuous text into independent structured clause units, each structured clause unit shall contain at least a unique clause number, complete clause text content, spatial location information, and adjacency and reference relationship markers with other structured clause units in the same document. S2.3. Perform boundary verification and error correction on the divided structured clause units. Through syntactic consistency detection and contextual coherence analysis, correct regional overlap, missing or misaligned areas caused by layout structure recognition deviations, and ensure the integrity and boundary accuracy of each structured clause unit, so as to provide accurate input for subsequent semantic parsing.

[0008] As a further improvement to this technical solution, in step S3, semantic parsing and semantic representation construction include the following steps: S3.1. Use a lightweight semantic parsing model that has been fine-tuned with civil aviation regulations corpus, or a large language model guided by the civil aviation field's special prompt word project, as the main body for semantic parsing execution; S3.2. Fine-grained semantic extraction is performed on the complete clause text content of each structured clause unit. The extracted content includes entity information in the civil aviation field, conditional relationships corresponding to conditional predicates, and obligation modalities corresponding to statutory norms. Among them, entity information in the civil aviation field includes airworthiness subject, aircraft type, operating scenario, compliance time limit, and responsible subject; conditional relationships include preconditions, triggering conditions, and exception conditions; and obligation modalities include mandatory obligations, prohibited negative obligations, and permitted authorized modalities. S3.3 Based on the extraction results, a standardized "entity-condition-modality" triple semantic representation is constructed for each structured clause unit. The triple semantic representation is the smallest execution unit of semantic parsing, providing standardized semantic input for subsequent construction of conditional dependencies.

[0009] As a further improvement to this technical solution, in step S3, performing semantic constraints and constructing conditional dependencies includes the following steps: S3.4. An ontology library for the civil aviation domain, constructed based on CAAC civil aviation regulations and standards and ICAO international civil aviation norms, is used to perform semantic constraints and consistency checks on the semantic representation of "entity-condition-modality" triples. The ontology library for the civil aviation domain includes the hierarchical relationships of entities in the civil aviation domain, attribute definitions, semantic association rules, and normative terminology mapping relationships. S3.5 Based on the semantic representation of "entity-condition-modality" triples that complete semantic constraints, identify the logical relationships between different structured clause units and construct conditional dependencies between clauses; among which conditional dependencies include premise-conclusion type causal dependencies and reference-referenced type normative dependencies; S3.6. Add a unique association identifier, dependency type label, and confidence score to each set of conditional dependencies to complete the standardized construction of conditional dependencies and provide standardized logical input for subsequent construction of conditional dependency graphs.

[0010] As a further improvement to this technical solution, in step S3, generating semantic evaluation information includes the following steps: S3.7. Construct a condition dependency graph using individual structured clause units as the set of nodes and standardized condition dependencies as the set of directed edges. The core formal definition is: ; in, For conditional dependency graphs, A set of nodes that corresponds one-to-one with a structured clause unit with a unique clause number. Let W be the set of directed edges representing the conditional dependencies between corresponding clauses, and let W be the set of edge weights corresponding to the confidence score of each conditional dependency. The range of values ​​for the inner element is [0,1]. S3.8 Based on the standard syntax of description logic and rule engine, the semantic representation of the "entity-condition-modality" triples of each node in the condition dependency graph and the condition dependency relationship of each directed edge are formally encoded, transforming the structured clause unit into a machine-executable logical unit and the condition dependency relationship into a logical jump relationship between structured clause units. S3.9. Based on the formally encoded logical units and logical jump relationships, construct a computable rule chain network. The specific verification process is as follows: In the condition dependency graph Extract simple directed paths with no repeating nodes as a single rule chain. All valid rule chains constitute a rule chain network; a single rule chain It covers two types of scenarios: complete logical paths and single-node independent paths; Calculate a single rule chain Completeness rating , The value range is [0,1]; based on A computability check was performed against a preset integrity threshold. Rule chains with a completeness threshold greater than or equal to the preset completeness threshold are considered valid rule chains, while those with a completeness threshold less than or equal to the preset completeness threshold are considered abnormal rule chains to be verified. The preset completeness threshold is a fixed configuration value within the closed interval [0,1]. S3.10, Node and Edge Attributes Based on Conditional Dependency Graphs The computability verification results are used to generate semantic evaluation information for the corresponding structured clause units.

[0011] As a further improvement to this technical solution, in step S4, logical consistency detection and identification of conditional dependency breaks between clauses and incomplete rule execution paths are performed, including the following steps: S4.1. Traverse all rule chains in the rule chain network and extract all abnormal rule chains to be verified as detection objects; S4.2. Combining the node attributes, edge attributes, and semantic evaluation information of the conditional dependency graph, identify the conditional dependency break type for each chain of abnormal rules to be verified; the specific identification rules are as follows: Node-missing type break: The judgment rule is that the preceding structured clause unit in the rule chain is missing, resulting in the preceding condition dependency relationship of the subsequent structured clause unit having no matching subject; Edge failure type fracture: The judgment rule is that the confidence score of the condition dependency relationship between the structured clause units in the rule chain is lower than the preset failure threshold; Incomplete path breakage: The judgment rule is the integrity score of the rule chain. Below the preset integrity threshold; Semantic conflict type break: The determination rule is that there is a semantic conflict in the semantic representation of the "entity-condition-modality" triple of the preceding and following structured clause units in the rule chain.

[0012] As a further improvement to this technical solution, in step S4, after identifying the condition-dependent fracture type, the rule chain is subjected to fracture quantification determination, including the following steps: S4.3 Construct a quantitative model for rule chain breakage and calculate the breakage score of a single rule chain. The calculation formula is: ; in, For the first The breakage score of the rule chain has a fixed value range of [0,1]. This represents the number of directed edges contained in the rule chain; S4.4. Set a preset fracture judgment threshold, which matches the fracture severity score range and is a fixed configuration parameter preset within the closed interval [0,1]. Based on the fracture severity score... Fracture determination is completed by comparing with the preset fracture determination threshold: when If the value is greater than or equal to the preset breakage judgment threshold, the rule chain is determined to be an abnormal rule chain with conditional breakage. when If the value is less than the preset breakage threshold, the rule chain is determined to be a valid rule chain without breakage. S4.5 For the abnormal rule chain that is determined to have condition-dependent fracture, associate it with the corresponding fracture type identified in S4.2, generate a fracture quantification judgment result containing fracture location, fracture type, and fracture degree score, and synchronously update the semantic evaluation information of the corresponding structured clause unit.

[0013] As a further improvement to this technical solution, in step S4, the broken or abnormal rule chains are completed or marked, and corresponding output information is generated, including the following steps: S4.6 Based on the fracture quantification judgment results, combined with semantic evaluation information and the civil aviation domain ontology library, semantic completion and repair are performed on the abnormal rule chains with conditional fractures. S4.7 For abnormal rule chains that cannot be semantically completed and repaired, mark the corresponding break type, break location and break degree score, form a fixed abnormal marker and include it in the semantic evaluation information. S4.8. Based on the condition dependency graph after completion or marking, generate rule execution path information; S4.9 Output standardized structured data and rule execution path information. The rule execution path information is used to support the analysis of the impact of clause revisions and the tracing of conditional dependencies.

[0014] As a further improvement to this technical solution, the data verification and dynamic update in S5 includes the following steps: S5.1. Conduct a comprehensive verification and quality assessment of the standardized structured data and rule execution path information output by S4. The verification content includes data integrity, semantic consistency, and rule execution path coherence. The quality assessment results are associated with the semantic assessment information of the corresponding structured clause unit. S5.2 Real-time monitoring of revision information of civil aviation regulations and manuals. When revision actions are detected, trigger full-process iterative updates from S1 to S4, and synchronously update the condition dependency graph, rule chain network, fracture quantification judgment results and semantic evaluation information to ensure that standardized structured data and rule execution path information are consistent with the latest civil aviation regulations and manuals.

[0015] The second objective of this invention is to provide a structured governance system for civil aviation regulations and manual data that integrates AI and OCR. The computer program running this system executes the aforementioned structured governance method for civil aviation regulations and manual data that integrates AI and OCR, including: The data acquisition and preprocessing module is used to preprocess the document images of civil aviation regulations and manuals, and extract the text data from the documents using OCR technology. The layout analysis and information location module, based on text data, identifies the layout structure of the document and divides the clause areas to obtain structured clause units; The semantic parsing and rule network construction module performs semantic parsing on structured clause units based on an artificial intelligence semantic parsing model, constructs a semantic representation containing entity information, conditional relationships, and obligation modalities; and applies semantic constraints to the semantic representation based on a civil aviation ontology library, constructs conditional dependencies between clauses, and then constructs a conditional dependency graph and a corresponding rule chain network based on the conditional dependencies, making the rule chain network computable and generating corresponding semantic evaluation information. The fracture detection and structured output module performs logical consistency detection on the rule chain network based on the condition dependency graph, identifies condition dependency fractures between clauses and incomplete rule execution paths; it also completes or marks fractured or abnormal rule chains based on the semantic evaluation results generated by the semantic parsing process; and finally generates rule execution path information based on the condition dependency graph, outputting standardized structured data and rule execution path information. The data verification and dynamic update module is used to verify and evaluate the quality of structured data and rule execution path information, and to trigger iterative updates of the data acquisition and preprocessing module, layout analysis and information positioning module, semantic parsing and rule network construction module, and fracture detection and structured output module based on the revision information of civil aviation regulations and manuals.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention integrates AI semantic parsing and OCR text extraction technologies. First, it preprocesses civil aviation regulations and manuals, analyzes their layout, and structurally breaks down clauses. Then, relying on the constraints of a civil aviation ontology library, it performs deep semantic parsing of the clauses, constructing clause entities, obligation modalities, and logical relationships. This leads to the establishment of refined conditional dependencies, conditional dependency graphs, and a computable rule chain network. Based on this, it performs logical consistency checks on the rule chain network, identifying logical gaps and dependency anomalies between clauses and intelligently completing or standardizing them. Simultaneously, it generates rule execution path information based on the improved conditional dependency graph and outputs standardized structured data. Furthermore, it can achieve full-process iterative updates and data quality verification based on civil aviation regulation revisions. This invention addresses the industry pain points of existing technologies, such as coarse modeling granularity, superficial clause logic characterization, and inability to support compliance verification and traceability of revision impacts. It enhances the professionalism, logic, and business adaptability of civil aviation regulations and operational manual data governance. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the method steps of the present invention; Figure 2 This is a schematic diagram of the overall process of the structured governance method in this invention; Figure 3 This is a schematic diagram of the system framework of the present invention; The meanings of the labels in the diagram are as follows: 1. Data acquisition and preprocessing module; 2. Page layout analysis and information location module; 3. Semantic parsing and rule network construction module; 4. Fracture detection and structured output module; 5. Data verification and dynamic update module. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figures 1-2 As shown, this embodiment provides a structured governance method for civil aviation regulations and manual data that integrates AI and OCR, including: S1. Data Acquisition and Preprocessing: Preprocess the images of civil aviation regulations and manuals, and extract the text data from the documents using OCR technology; specifically including the following processes: First, obtain the civil aviation regulations document or operation manual document to be processed. The document can be in PDF format or image format. When it is in PDF format, convert it into image data page by page, and perform resolution uniform processing on all images. Preferably, the resolution is set to 300dpi to improve the accuracy of subsequent text recognition.

[0020] Secondly, the image data is preprocessed, including: converting the image to grayscale to reduce the dimensionality of the image data; using an adaptive thresholding method to binarize the image to enhance the contrast between the text and the background; performing tilt detection on the text lines in the image based on Hough transform, and rotating the image when a tilt angle is detected, with the correction angle preferably within ±15°; and simultaneously, performing median filtering on the image to remove scanning noise and speckle interference, thereby improving the clarity of character boundaries.

[0021] After preprocessing, the processed image is input into the OCR recognition process to extract text information from the image. During this process, not only is the text content acquired, but also the spatial location information corresponding to the text is simultaneously obtained, including the boundary coordinates of text blocks and their distribution on the page.

[0022] Furthermore, the text data obtained from OCR recognition is standardized, including character unification (e.g., converting full-width characters to half-width characters), punctuation standardization, and filtering of abnormal characters. At the same time, by statistically analyzing the frequency of repeated occurrences of text on each page, repetitive content such as headers, footers, and page numbers is identified and removed to reduce the impact of redundant information on subsequent processing.

[0023] After the above processing, structured text data containing text content and its spatial location information is obtained, which serves as input data for subsequent steps of layout analysis and semantic parsing.

[0024] S2. Page Layout Analysis and Information Location: Based on text data, the document's page layout structure is identified and clause areas are divided to obtain structured clause units; In this embodiment, the layout analysis and information location of S2 includes the following steps: S2.1 Based on text data, layout structure recognition technology is used to identify structural elements in the document and generate layout metadata containing position coordinates and hierarchical relationships; among which structural elements include heading levels, chapter numbers, clause numbers, paragraph boundaries, table structures, and appendix markers, etc. Specifically, the text content and its spatial location information obtained in step S1 are used as input, and the structural elements in the document are identified by combining the coordinate distribution of text blocks, font features (including font size and bolding), and numbering patterns. Numbering patterns include, but are not limited to, regular formats such as "Chapter X", "XX", and "XXX".

[0025] Furthermore, during the identification process, different structural elements are judged by combining preset rules with models, wherein: When text meets the preset title characteristics (such as font size larger than the body text, located at the beginning of a line, with numbering or keywords), it is determined to be a title level; When text conforms to the chapter numbering rules, it is identified as a chapter number; When the text contains a clause number (such as "3.2.1" or "(a)"), it is identified as the beginning of the clause; Determine paragraph boundaries based on line spacing and alignment relationships between text blocks; For text regions with a regular grid distribution, they are identified as table structures; Areas containing keywords such as "Appendix" or "Attachment" are identified as appendix tags.

[0026] Finally, after completing the above identification, corresponding page metadata is generated for each structural element. The page metadata includes at least: structural element type, page number, boundary coordinates (x1, y1, x2, y2), and hierarchical relationship information. The hierarchical relationship is automatically constructed based on the title number, such as the nested relationship of chapter-section-article.

[0027] S2.2. Use page layout metadata to divide the document into clause areas, cut continuous text into independent structured clause units, each structured clause unit shall contain at least a unique clause number, complete clause text content, spatial location information, and adjacency and reference relationship markers with other structured clause units in the same document. Specifically, after obtaining the page layout metadata, the continuous text in the document is segmented using the clause number as the primary dividing criterion. Specifically, starting from the identified clause number, the segment extends to the next clause number or the location of a sibling element, forming a complete clause area.

[0028] During the segmentation process, multiple paragraphs within the same clause are merged based on paragraph boundary information to ensure the integrity of the clause content; for clauses spanning multiple pages, they are spliced ​​together based on page continuity and numbering consistency.

[0029] After division, each structured clause unit should contain at least the following information: Unique clause number (used to identify this clause unit); The complete text of the terms and conditions (including the full text of the relevant paragraphs); Spatial location information (including the coordinate range and page number information of the clause in the original document); Adjacent relationships (including the order of preceding and subsequent clauses); Reference relationship marker (when the clause contains expressions such as "see Article X", establish the association between the referenced clause and the cited clause).

[0030] S2.3. Perform boundary verification and error correction on the divided structured clause units. Through syntactic consistency detection and contextual coherence analysis, correct regional overlap, missing or misaligned areas caused by layout structure recognition deviations, and ensure the integrity and boundary accuracy of each structured clause unit, so as to provide accurate input for subsequent semantic parsing.

[0031] Specifically, the verification and correction process includes: First, a syntactic consistency check is performed. The text in each clause unit is analyzed for sentence integrity. When semantic breaks, paragraphs are cut across clauses, or there is no closing number, it is determined that the clause unit may have a truncation problem, and it is expanded or merged to adjacent text areas for correction.

[0032] Secondly, a contextual coherence analysis is performed. By calculating the semantic similarity between adjacent clause units (preferably using a similarity calculation method based on vector representation), when the semantic continuity between adjacent texts is lower than a preset threshold (preferably 0.7-0.8), it is determined that there is a division deviation, and the clause boundaries are adjusted to eliminate regional misalignment or content breakage caused by page structure recognition errors.

[0033] At the same time, spatial relationship verification is performed on all clause units to detect whether there are overlapping areas or abnormal gaps; when there is overlap, the area is re-divided according to the clause number priority or text integrity; when there is a missing part, the missing part is supplemented by adjacency relationship.

[0034] Through the above verification and error correction processes, we ensure that each structured clause unit is semantically complete and spatially accurate, thereby providing reliable and accurate input data for subsequent semantic parsing steps.

[0035] S3. Semantic parsing and rule network construction: The structured clause unit is semantically parsed using an AI-based semantic parsing model to construct a semantic representation containing entity information, conditional relationships, and obligation modalities. The semantic representation is semantically constrained based on the civil aviation ontology library to construct conditional dependencies between clauses. Based on the conditional dependencies, a conditional dependency graph and the corresponding rule chain network are constructed to make the rule chain network computable and generate corresponding semantic evaluation information. In this embodiment, step S3, performing semantic parsing and constructing semantic representation, includes the following steps: S3.1. Use a lightweight semantic parsing model that has been fine-tuned with civil aviation regulations corpus, or a large language model guided by the civil aviation field's special prompt word project, as the main body for semantic parsing execution; Specifically, the semantic parsing process can be implemented using one of the following two methods or a combination thereof: One approach involves employing a lightweight semantic parsing model based on a pre-trained language model. Specifically, a general pre-trained model is used as a foundation, and a civil aviation regulations corpus is introduced for fine-tuning and training. The training corpus preferably includes no fewer than 100,000 civil aviation regulations and manual clauses. During training, entity recognition, conditional extraction, and modality classification tasks are constructed using labeled data, enabling the model to perform specialized semantic parsing for civil aviation texts. To improve processing efficiency, the model parameter size is preferably controlled within the millions (≤100M) to balance inference speed and parsing accuracy.

[0036] Another approach involves using a large language model, guided by a civil aviation-specific prompting term project. Specifically, standardized prompting term templates are constructed, the clause text is input into the model, and the model is guided to output structured semantic information. For example, prompting terms explicitly require the model to extract elements such as "subject, condition, and type of obligation," thereby obtaining semantic parsing results that conform to a preset format.

[0037] In practical applications, any of the above methods can be selected according to computing resources and accuracy requirements, or a lightweight model can be used for preliminary analysis first, and then a large language model can be used for supplementation or correction to improve the overall analysis quality.

[0038] By setting the semantic parsing execution entity as described above, the semantic information of the clauses can be extracted stably and accurately, providing basic support for the subsequent construction of conditional dependencies and the generation of rule chains.

[0039] S3.2. Fine-grained semantic extraction is performed on the complete clause text content of each structured clause unit. The extracted content includes entity information in the civil aviation field, conditional relationships corresponding to conditional predicates, and obligation modalities corresponding to statutory norms. Among them, entity information in the civil aviation field includes airworthiness subject, aircraft type, operating scenario, compliance time limit, and responsible subject; conditional relationships include preconditions, triggering conditions, and exception conditions; and obligation modalities include mandatory obligations, prohibited negative obligations, and permitted authorized modalities. In this step, after determining the subject responsible for semantic parsing, fine-grained semantic extraction is performed on the complete clause text content of each structured clause unit obtained in step S2 to obtain computable core semantic elements.

[0040] Specifically, the text of each clause is input into a semantic parsing model. Through a combination of sequence labeling and semantic role labeling, the clause is parsed word-by-word or phrase-by-phrase, and then categorized and extracted according to preset semantic categories. The extracted content includes entity information, conditional relationships, and obligation modalities in the civil aviation field, among which: Entity information extraction in the civil aviation sector: Based on the named entity recognition method, key entities are identified and extracted from the terms and mapped to a predefined set of entity categories, which include at least: Airworthiness entities: such as "airline", "maintenance unit", etc.; Models: such as "B737", "A320", etc.; Operational scenarios: such as "before takeoff", "cruise phase", "emergency situations", etc. Compliance time limits: such as time constraints expressed as "within 30 minutes before takeoff" or "during flight"; Responsible parties include: such as "captain", "first officer", and "dispatcher".

[0041] During the extraction process, different expressions of the same entity are uniformly identified by combining contextual semantics. For example, "captain" and "commander captain" are grouped into the same entity category.

[0042] Conditional relation extraction: Based on dependency parsing and trigger word identification, conditional predicate structures are identified from the clauses, and corresponding conditional relations are extracted. These conditional relations include at least: Prerequisites: Describe the basic premises under which the rule applies, such as "before flight" or "when weather conditions are met"; Triggering conditions: Describe the specific event that activates the rule, such as "when the system fails" or "if a deviation from the flight path occurs"; Exceptions: Describe situations where the rule does not apply or is exempted, such as "unless approved" or "except in emergency situations".

[0043] In practice, the condition range is determined by identifying conditional trigger words such as "when," "if," "under certain circumstances," and "unless," combined with syntactic structure, and the complete condition expression is extracted.

[0044] Obligation modality extraction: Based on modal verb recognition and semantic classification methods, normative expressions in clauses are identified and classified into different obligation types, including at least: Mandatory obligations of the "must" category: such as "must", "shall", "must", etc.; Prohibited negative obligations: such as "must not" or "prohibited"; Optional authorization obligations: such as "may", "permit", etc.

[0045] During the extraction process, by constructing a modal dictionary and combining it with contextual semantics, misjudgments caused by ambiguous expressions are avoided, such as distinguishing the different semantics of "can be used for explanation" and "can be used to perform operations".

[0046] Through the fine-grained semantic extraction described above, the original natural language clauses are decomposed into a set of semantic elements with a clear structure, laying the foundation for subsequent formal expression.

[0047] S3.3 Based on the extraction results, a standardized "entity-condition-modality" triple semantic representation is constructed for each structured clause unit. The triple semantic representation is the smallest execution unit of semantic parsing, providing standardized semantic input for subsequent construction of conditional dependencies.

[0048] In this step, after completing fine-grained semantic extraction, the extraction results are structured and modeled to construct a unified triplet semantic representation for each structured clause unit.

[0049] Specifically, entity information, conditional relationships, and obligation modalities within the same clause are combined according to preset rules to construct "entity-condition-modality" triples, in the following form: Entity: Represents the object or subject of a rule's action; Condition: Indicates the prerequisites for a rule to be true or triggered; Modality: Represents the constraint type of a rule.

[0050] During the construction process: First, the extracted entities are filtered to determine the core entity as the subject of the triple, and the remaining entities are used as secondary attributes or supplementary contextual information. Secondly, multiple conditional relationships are combined according to logical relationships. When there are multiple conditions, a compound conditional expression is constructed based on their logical relationships, such as using "AND" or "OR" to connect them. Finally, the obligation modality is treated as a constraint attribute of the triple, which together with the entity and condition constitutes a complete semantic unit.

[0051] For example, the clause "The captain must complete the flight control check before takeoff" can be represented by the following triple: Entity: Captain; Condition: Before takeoff; Modal: Must be executed; Furthermore, to enhance computability, triples are standardized and encoded, converted into a unified data structure, and each triple is assigned a unique identifier.

[0052] Triple semantic representations, as the smallest unit of execution in semantic parsing, possess the following characteristics: Semantic integrity: The ability to independently express a rule constraint; Structural uniformity: Facilitates cross-clause comparison and calculation; Composability: Conditional dependencies and rule chains can be further constructed.

[0053] By constructing the above-mentioned standardized triple semantic representation, a unified and computable semantic input is provided for subsequent conditional dependency modeling and rule chain network generation, which is a key foundation for realizing automatic reasoning and consistency analysis of regulations.

[0054] In this embodiment, step S3, performing semantic constraints and constructing conditional dependencies, includes the following steps: S3.4. An ontology library for the civil aviation domain, constructed based on CAAC civil aviation regulations and standards and ICAO international civil aviation norms, is used to perform semantic constraints and consistency checks on the semantic representation of "entity-condition-modality" triples. The ontology library for the civil aviation domain includes the hierarchical relationships of entities in the civil aviation domain, attribute definitions, semantic association rules, and normative terminology mapping relationships. In this step, the entity information in the triples is subjected to normalization mapping to eliminate semantic ambiguity caused by different representations. The mapping process is as follows: ; in: Represents the standardized entity; Represents the original extracted entity; This represents a set of entities in the civil aviation ontology database; Represents candidate entities in the ontology library; This represents the semantic similarity function.

[0055] The above methods achieve unified and consistent entity representation at the ontology level.

[0056] In addition, based on the attribute constraint rules in the civil aviation ontology library, consistency checks are performed on the conditions and modalities, including: Does the condition fall under a legal operating scenario? Does the modality match the authority of the responsible party? If the constraint rules are not met, the corresponding triples will be marked as conflicting or removed.

[0057] S3.5 Based on the semantic representation of "entity-condition-modality" triples that complete semantic constraints, identify the logical relationships between different structured clause units and construct conditional dependencies between clauses; among which conditional dependencies include premise-conclusion type causal dependencies and reference-referenced type normative dependencies; In this step, the dependency determination function between clauses is defined as follows:

[0058] in: Terms and Conditions With Terms Dependency markers between them; This indicates the triples corresponding to different clauses; This represents the dependency decision function.

[0059] Furthermore, the dependency decision function introduces a multi-source decision mechanism, including: Determining the inclusion relationship: ; indicates terms The conditions are the terms A subset of the conditions; Semantic similarity triggers the judgment: ; Standardized citation relationship identification (based on text trigger word rules); in, This represents the set of conditions corresponding to the clause; This represents the similarity threshold.

[0060] When any one of the conditions is met, it is considered that there is a dependency relationship between the clauses.

[0061] S3.6. Add a unique association identifier, dependency type label, and confidence score to each set of conditional dependencies to complete the standardized construction of conditional dependencies and provide standardized logical input for subsequent construction of conditional dependency graphs.

[0062] In this step, the confidence score is the core calculation process, and the following formula is used: ; in: Terms and Conditions and Confidence score of the dependency relationship between them; , Denotes the weight coefficients, and satisfies ; Indicates the semantic similarity of the terms; This indicates the probability of predicting dependencies.

[0063] This implementation provides the following calculation example: set up: ; ; ; but: ; This indicates that the dependency relationship of this clause has a high degree of credibility.

[0064] Furthermore, generate a unique identifier for each dependency: ; in: This represents a unique identifier for the dependency relationship; This represents a hash function.

[0065] The final result is a standardized relational structure containing dependency identifiers, dependency types, and confidence scores, which is used for subsequent construction of conditional dependency graphs.

[0066] In this embodiment, generating semantic evaluation information in step S3 includes the following steps: S3.7. Construct a condition dependency graph using individual structured clause units as the set of nodes and standardized condition dependencies as the set of directed edges. The core formal definition is: ; in, For conditional dependency graphs, A set of nodes that corresponds one-to-one with a structured clause unit with a unique clause number. This is the set of directed edges representing the conditional dependencies between corresponding clauses. Let be the set of edge weights corresponding to the confidence score of each conditional dependency, and The range of values ​​for the inner element is [0,1]. In this step, the specific implementation process for constructing the condition dependency graph is as follows: First, a node set is established, with each structured clause unit as a graph node. Each node uniquely corresponds to a triplet representation that has completed semantic constraints.

[0067] Secondly, iterate through all dependency pairs. When satisfied At the node With nodes Establish a directed edge between them The direction of the edge points from the premise clause to the conclusion clause.

[0068] Subsequently, the confidence score calculated in step S3.6 is... Assign the value to the corresponding edge The weights are used to characterize the reliability of dependencies.

[0069] Finally, the conditional dependency graph is constructed: ; The discrete clause relationships are transformed into a unified directed weighted graph structure using the above method.

[0070] S3.8 Based on the standard syntax of description logic and rule engine, the semantic representation of the "entity-condition-modality" triples of each node in the condition dependency graph and the condition dependency relationship of each directed edge are formally encoded, transforming the structured clause unit into a machine-executable logical unit and the condition dependency relationship into a logical jump relationship between structured clause units. In this step, the "entity-condition-modality" triple for each node is converted into a logical expression: ; in: Indicates the first The logical unit corresponding to each clause; Represents standardized entities; Represents a set of conditions; This represents the constraint-driven action based on the obligation modality. For the duty modal type, This refers to the civil aviation compliance operation actions corresponding to this mode.

[0071] The implementation method is as follows: condition set Convert to a Boolean expression (AND / OR structure); Map the obligatory modalities to corresponding constraint actions, such as mapping mandatory modalities to enforced actions, prohibited modalities to stop / avoid actions, and permitted modalities to optional actions; Generate a unified rule format (such as the IF-THEN structure).

[0072] Furthermore, for each directed edge Construct logical jump rules: ; in: Each represents a separate logical unit; This indicates the rule triggering relationship.

[0073] The implementation method is as follows: Current logic unit When the condition is met, the subsequent logic unit is triggered. ; Transform the edges in the graph structure into rules execution order constraints.

[0074] Through the above processing, the transformation from "graph structure" to "executable rules" is achieved.

[0075] S3.9. Based on the formally encoded logical units and logical jump relationships, construct a computable rule chain network. The specific verification process is as follows: In the condition dependency graph Extract simple directed paths with no repeating nodes as a single rule chain. All valid rule chains constitute a rule chain network; a single rule chain It covers two types of scenarios: complete logical paths and single-node independent paths; Calculate a single rule chain Completeness rating , The value range is [0,1]; based on A computability check was performed against a preset integrity threshold. Rule chains with a completeness threshold greater than or equal to the preset completeness threshold are considered valid rule chains, while those with a completeness threshold less than or equal to the preset completeness threshold are considered abnormal rule chains to be verified. The preset completeness threshold is a fixed configuration value within the closed interval [0,1]. In this step, firstly, depth-first search (DFS) combined with strongly connected components (SCC) is used to detect the conditional dependency graph. Perform path traversal: Extract simple directed paths without repeating nodes as standard rule chains; Identify and preserve strongly connected subgraphs with cyclic dependencies, and mark them as cyclic rule chains; all paths satisfy the following conditions: directed path, length greater than or equal to 1; Obtain the rule chain: ; in: Indicates the first A chain of rules; This indicates the number of nodes in the rule chain (i.e., the length of the rule chain, which is the number of clause units).

[0076] Furthermore, for each rule chain, its completeness score is calculated: ; in: This indicates the score for the completeness of the rule chain; Indicates the number of nodes in the rule chain; Indicates the first rule in the chain. The structured clause unit and the first The confidence score corresponds to the conditional dependencies between structured clause units.

[0077] In addition, this embodiment also provides the following calculation example: Suppose a rule chain contains 4 nodes, and its edge weights are as follows: ; ; ; but: ; When the threshold is set to 0.75, the rule chain is determined to be a valid rule chain.

[0078] Finally, the following determination mechanism is set: like Mark it as a valid rule chain; like Marked as a low-confidence rule chain; in, This is a preset integrity threshold.

[0079] S3.10, Node and Edge Attributes Based on Conditional Dependency Graphs The computability verification results are used to generate semantic evaluation information for the corresponding structured clause units.

[0080] Specifically, the structured clause units are first traversed one by one based on the nodes in the condition dependency graph. For each node, its participation in the rule chain network is considered: if the node appears in at least one rule chain that has been determined to be valid, the clause is considered to have a clear role in the overall logical structure and is marked as a valid clause; if the node does not participate in any valid rule chain, it is marked as an isolated clause to indicate that it has not formed a valid linkage relationship in the current dependency structure.

[0081] After completing the node-level determination, the edges in the conditional dependency graph are further processed. Specifically, the corresponding edge is read. The dependency is then compared with a preset threshold. If the weight is greater than or equal to the threshold, the dependency is marked as a high-confidence dependency; otherwise, it is marked as a weak dependency. This method directly distinguishes between stable and uncertain dependencies, facilitating subsequent screening and verification.

[0082] Subsequently, the rule chain hierarchy is summarized. This involves applying the completeness score calculated in step S3.9. The rule chain is configured, and its integrity status is recorded for each rule: when the score meets the threshold condition, the rule chain is marked as a complete rule chain; otherwise, it is marked as an incomplete rule chain. Incomplete rule chains are also individually identified for focused inspection or supplementary analysis in subsequent processing.

[0083] Based on the above processing, all results are organized into a unified structure. Specific output includes: The validity marking results for each structured clause unit; Conditional dependencies between clauses and their credibility classification results; The set of rule chains and their corresponding integrity scores and integrity status; By using the above method, the calculation results of the condition dependency graph and rule chain are transformed into structured evaluation results with clearly marked information, so that the logical relationships between the clauses and their reliability can be directly identified and used.

[0084] S4. Break Detection and Structured Output: Based on the condition dependency graph, logical consistency detection is performed on the rule chain network to identify breaks in condition dependencies between clauses and incomplete rule execution paths; broken or abnormal rule chains are completed or marked based on the semantic evaluation results generated during the semantic parsing process; rule execution path information is generated based on the condition dependency graph; standardized structured data and rule execution path information are output, which are used to support the analysis of the propagation of the impact of clause revisions and the tracing of condition dependency paths; In this embodiment, step S4, which involves logical consistency detection and identifying breaks in conditional dependencies between clauses and incomplete rule execution paths, includes the following steps: S4.1. Traverse all rule chains in the rule chain network and extract all abnormal rule chains to be verified as detection objects; In this step, after completing the rule chain extraction and integrity calculation, the rule chain network is traversed, and the rule chains are filtered in combination with the rule chain integrity judgment results obtained in step S3.

[0085] Specifically, the rule chains marked as incomplete rule chains in step S3 are extracted and used as abnormal rule chains to be verified. At the same time, rule chains containing low-confidence dependencies are also included in the scope of verification.

[0086] The determination of incomplete rule chains and low-confidence dependencies is based on the completeness score obtained in the preceding steps. and The comparison results are confirmed.

[0087] Ultimately, a set of abnormal rule chains to be verified is formed, which serves as the input object for subsequent fracture detection.

[0088] S4.2. Combining the node attributes, edge attributes, and semantic evaluation information of the conditional dependency graph, identify the conditional dependency break type for each chain of abnormal rules to be verified; the specific identification rules are as follows: Node-missing type break: The judgment rule is that the preceding structured clause unit in the rule chain is missing, resulting in the preceding condition dependency relationship of the subsequent structured clause unit having no matching subject; Edge failure type fracture: The judgment rule is that the confidence score of the condition dependency relationship between the structured clause units in the rule chain is lower than the preset failure threshold; Incomplete path breakage: The judgment rule is the integrity score of the rule chain. Below the preset integrity threshold; Semantic conflict type break: The determination rule is that there is a semantic conflict in the semantic representation of the "entity-condition-modality" triple of the preceding and following structured clause units in the rule chain.

[0089] In this step, for the set of abnormal rule chains to be verified obtained in step S4.1, the rule chains are analyzed one by one in combination with the node relationships, dependency weights and semantic evaluation results in the condition dependency graph, in order to identify different types of condition dependency breaks.

[0090] Before performing fracture identification, the relevant thresholds involved in the judgment process are first set: the sample quantile statistical method is adopted, and the distribution quantiles of edge weights and integrity scores are calculated based on the historical civil aviation regulation rule chain sample set. The dependency failure threshold is taken as the lower quartile of the weight distribution, and the integrity threshold is taken as the median of the integrity distribution of the effective rule chain, to ensure that the thresholds have statistical significance and engineering reproducibility.

[0091] In one exemplary implementation: The dependency failure threshold can be determined based on the quantiles of the edge weights, for example, by taking all edges. The lower quartile of the distribution corresponds to a numerical range of 0.3 to 0.5; The rule chain integrity threshold can be determined based on the average level of valid rule chain samples, for example, taking 0.6 to 0.8 as the judgment range.

[0092] The above thresholds can be adjusted in practice based on the rule size and domain differences.

[0093] After setting the threshold, each rule chain is traversed in node order, and the break type is identified according to the following rules: Identification of missing node fractures: For any subsequent structured clause unit in the rule chain, retrieve its set of preceding dependent nodes in the condition dependency graph, and match this set with existing preceding nodes in the rule chain. If no preceding node that can form a valid dependency relationship is found, it is determined to be a node missing type break.

[0094] This situation typically manifests as a lack of necessary prerequisites in subsequent clauses of the rule chain, resulting in the inability to establish a dependency relationship.

[0095] Edge failure type fracture identification: For the dependency relationship between any adjacent nodes in the rule chain, read the corresponding... The relationship is then compared with a preset dependency failure threshold. If the threshold is lower than the threshold, the dependency is determined to be a failed dependency, and the rule chain is marked as having an edge failure type break.

[0096] This type of breakage reflects that although the dependency exists, its reliability is insufficient and it cannot effectively support the stable execution of the rule chain.

[0097] Identification of incomplete path fractures: For the rule chain as a whole, read its completeness score. It is then compared with a preset integrity threshold. When it falls below the threshold, the rule chain is determined to be a path incompleteness type break.

[0098] This situation indicates that the rule chain has structural breakpoints or insufficient dependency connections, and does not have a complete execution path as a whole.

[0099] Semantic conflict-based break detection: For structured clause units that are adjacent or have dependencies in the rule chain, extract their corresponding "entity-condition-modality" semantic representations and perform consistency comparisons. A semantic conflict-type break is determined when any of the following conditions are met: The objects being described are the same, but the constraints or modal expressions are contradictory; The conditions are mutually exclusive, which means that the rules cannot all be true at the same time; It is marked as an inconsistent relation in the semantic evaluation results.

[0100] This type of break reflects a semantic conflict in the rule chain, causing the logic to fail.

[0101] Finally, after completing the above identification, the break type and corresponding break location of each rule chain are recorded to provide a basis for subsequent processing.

[0102] In this embodiment, after identifying the condition-dependent break type in step S4, the rule chain is quantified and determined, including the following steps: S4.3 Construct a quantitative model for rule chain breakage and calculate the breakage score of a single rule chain. The calculation formula is: ; in, For the first The breakage score of the rule chain has a fixed value range of [0,1]. This represents the number of directed edges contained in the rule chain.

[0103] S4.4. Set a preset fracture judgment threshold, which matches the fracture severity score range and is a fixed configuration parameter preset within the closed interval [0,1]. Based on the fracture severity score... Fracture determination is completed by comparing with the preset fracture determination threshold: when If the value is greater than or equal to the preset breakage judgment threshold, the rule chain is determined to be an abnormal rule chain with conditional breakage. when If the value is less than the preset breakage threshold, the rule chain is determined to be a valid rule chain without breakage. In this step, the preset fracture judgment threshold is a fixed configuration parameter within the closed interval [0,1], which matches the value range of the fracture score. It can be preset according to the structured governance needs of different civil aviation regulatory texts. For example, in the compliance verification scenario for civil aviation operation regulations, the preset fracture judgment threshold can be set to 0.3 to distinguish the fracture risk level of the rule chain.

[0104] Fracture detection logic: when When the value is greater than or equal to the preset breakage threshold, the rule chain is determined to be an abnormal rule chain with conditional breakage. when When the preset breakage threshold is reached, the rule chain is determined to be a valid rule chain without breakage.

[0105] S4.5 For the abnormal rule chain that is determined to have condition-dependent fracture, associate it with the corresponding fracture type identified in S4.2, generate a fracture quantification judgment result containing fracture location, fracture type, and fracture degree score, and synchronously update the semantic evaluation information of the corresponding structured clause unit.

[0106] In this step, for rule chains identified as abnormal in S4.4, they are first precisely associated with the break types identified in S4.2. The association process is based on the node composition, directed edge weights, and semantic conflict detection results of the rule chains, and matches one by one the four break types: node missing type, edge invalid type, path incomplete type, and semantic conflict type, to clarify the core break cause and specific location of each abnormal rule chain.

[0107] After completing the association of fracture types, a standardized fracture quantification judgment result is generated for each abnormal rule chain. This result is recorded in a structured form, including five core contents: unique rule chain identifier, fracture location, fracture type, fracture degree score, and confidence score of the involved dependency edge. All data are taken from the actual calculation results of the aforementioned steps, without any new or fabricated information.

[0108] After generating the fracture quantification judgment result, the above information is synchronously updated to the semantic evaluation information of the corresponding structured clause unit and rule chain. The update operation is divided into three dimensions: First, fracture association markers are added to the preceding and following clause nodes involved in the fracture location, clearly indicating the abnormal rule chain in which the clause participates, the fracture type, and the fracture risk warning; second, a fracture quantification judgment sub-item is added to the overall semantic evaluation information of the rule chain, fully recording all content such as fracture location, type, and score; third, the attributes of the corresponding failure edge in the condition dependency graph are marked, and its confidence status is synchronously updated to the semantic evaluation system. The updated semantic evaluation information can be directly used for subsequent clause semantic completion and abnormal rule chain marking, while providing complete data support for the impact propagation analysis and condition dependency path tracing during the revision of civil aviation regulations.

[0109] For valid rule chains determined by S4.4 to be without breaks, and rule chains with independent paths at single nodes, there is no need to perform break type association and result generation operations. Their semantic evaluation information remains in its original state, retaining only basic evaluation content such as clause validity and dependency confidence, ensuring the simplicity and effectiveness of the overall data processing. Through the above complete and standardized processing flow, accurate recording of break information and dynamic improvement of semantic evaluation information are achieved, ensuring the openness and transparency of the structured governance process of civil aviation regulations and the authenticity and reliability of data.

[0110] In this embodiment, in step S4, the broken or abnormal rule chains are completed or marked, and corresponding output information is generated, including the following steps: S4.6 Based on the fracture quantification judgment results, combined with semantic evaluation information and the civil aviation domain ontology library, semantic completion and repair are performed on the abnormal rule chains with conditional fractures. Specifically, the fracture quantification judgment result of the abnormal rule chain is retrieved first to clarify the fracture type, fracture location and fracture degree score. Then, the semantic evaluation information of the structured clause units involved in the rule chain is matched to obtain the semantic representation of the condition modal triple of each clause entity, the condition dependency confidence and semantic consistency status.

[0111] Subsequently, based on the civil aviation database, targeted completion and repair were carried out for different types of fractures: For node-missing breaks, based on the entity and condition constraints of the subsequent clauses, the missing preceding dependency nodes are marked in the condition dependency diagram. The original regulatory text is retained without any additions, and only the location of the dependency break is marked.

[0112] For edge failure-type fractures, based on the semantic association rules of the ontology, the condition matching relationship of low-confidence dependent edges is corrected, the dependency confidence is calibrated to the effective range, and the validity of the edges is restored. For path incompleteness-type breaks, according to the logical paradigm of the ontology library, the missing intermediate condition dependencies of the rule chain are supplemented to improve the complete logical link of rule execution. For semantically conflict-type breaks, the semantic representation of the triples of the conflicting clauses is adjusted according to the semantic constraint verification rules of the ontology library to resolve the contradictions between entities, conditions or modalities and restore the logical consistency of the rule chain.

[0113] The entire process of supplementing and repairing strictly followed the original clause text and civil aviation regulations, without altering the core semantics of the clauses, and only repaired logical breaks.

[0114] S4.7 For abnormal rule chains that cannot be semantically completed and repaired, mark the corresponding break type, break location and break degree score, form a fixed abnormal marker and include it in the semantic evaluation information. Specifically, for abnormal rule chains that cannot be repaired even after completion by S4.6, an objective marking operation is performed. The marking content is directly taken from the fracture quantification judgment result, accurately marking the three core pieces of information of the rule chain: fracture type, specific fracture location, and fracture degree score. The marking establishes a unique association with the rule chain and the structured clause unit involved, avoiding information confusion.

[0115] After marking is completed, the fixed anomaly markers are synchronously incorporated into the semantic evaluation information of the corresponding structured clause unit and rule chain, and permanently retained as a fixed component of the semantic evaluation. The anomaly markers only objectively record the break state and quantitative data, without adding subjective judgment content, which facilitates the rapid location of the root cause of the anomaly during subsequent manual review and clause revision.

[0116] S4.8. Based on the condition dependency graph after completion or marking, generate rule execution path information; Specifically, based on the condition dependency graph repaired by S4.6 and marked by S4.7, all nodes and directed edges in the graph are traversed to extract the complete execution link of each rule chain and generate rule execution path information.

[0117] The path information includes four core components: the structured clause unit number involved in the path, the conditional dependencies between clauses, the confidence score of the dependency edges, and the rule chain status (fixed / anomaly marked). For valid rule chains that have been completed, the path clearly presents a complete and executable logical link; for rule chains marked as anomaly, the path synchronously carries the anomaly mark, clearly indicating the location of broken nodes and failed edges.

[0118] All path information is based solely on the processed condition dependency graph, fully preserving the logical structure and state information, and can be directly used for subsequent path tracing and compliance verification.

[0119] S4.9 Output standardized structured data and rule execution path information. The rule execution path information is used to support the analysis of the impact of clause revisions and the tracing of conditional dependencies.

[0120] Specifically, the structured clause units, entity-condition-modal triple semantic data, and condition dependency relationship data processed throughout the entire process are integrated into standardized structured data in a unified format, and then bound with the rule execution path information to form the final output dataset.

[0121] The output rule execution path information serves two core purposes: first, it supports the analysis of the propagation of the impact of clause revisions. When a civil aviation regulation clause is revised, the path information can quickly locate all related rule chains and dependent clauses, clarifying the scope of the revision's impact; second, it supports conditional dependency path tracing. Starting from any target clause, it can reverse-engineer all preceding dependent clauses and the complete execution chain, reconstructing the logical constraint relationships between clauses. The output content contains only verified and valid data, with no redundant information, and can be directly applied to practical scenarios such as structured governance and compliance management of civil aviation regulations and manuals.

[0122] S5. Data Validation and Dynamic Update: Validate and assess the quality of structured data and rule execution path information, and trigger iterative updates of steps S1 to S4 based on revisions to civil aviation regulations and manuals.

[0123] In this embodiment, the data verification and dynamic update in S5 includes the following steps: S5.1 A comprehensive verification and quality assessment is performed on the standardized structured data and rule execution path information output by S4. Verification includes data integrity, semantic consistency, and rule execution path coherence. The quality assessment results are correlated with the semantic assessment information of the corresponding structured clause units. Data integrity verification focuses on checking whether the structured clause units completely cover all valid clauses in the original document, without omissions or truncation; whether basic information such as the number, text, location, and reference relationships of individual clauses are complete; and whether semantic structures such as entity-condition-modal triples, conditional dependencies, and rule chains are free of empty items or incompleteness.

[0124] Semantic consistency verification uses the civil aviation ontology database as the sole benchmark, checking item by item whether the expression of the clause entity, condition, and modality conforms to the CAAC and ICAO standard terminology, whether there are semantic conflicts in the dependencies between clauses, and whether the matching of the obligation modality with the responsible subject and operational scenario conforms to the logic of civil aviation regulations, to ensure that the semantics are unambiguous and without contradictions.

[0125] The rule execution path continuity verification mainly checks: whether the execution link of the repaired rule chain is smooth and without breaks; whether the rule chains marked as abnormal accurately carry break indicators; and whether the path nodes, dependency edges, and condition dependency graphs correspond completely, with no path confusion or association failure issues. After completing all verifications, a quality assessment result is generated for each structured clause unit and each rule chain. The assessment content is associated with and synchronously stored with the semantic assessment information of the corresponding clause, serving as the basis for data quality identification and subsequent review and application.

[0126] S5.2 Real-time monitoring of revision information of civil aviation regulations and manuals. When revision actions are detected, trigger full-process iterative updates from S1 to S4, and synchronously update the condition dependency graph, rule chain network, fracture quantification judgment results and semantic evaluation information to ensure that standardized structured data and rule execution path information are consistent with the latest civil aviation regulations and manuals.

[0127] Specifically, the system adopts a combination of timed comparison and proactive verification to track the differences between the current document version on which governance is based and the latest released version in real time. When revision actions such as adding clauses, modifying content, or deleting items are detected, the system automatically triggers a full-process iterative update from S1 data collection and preprocessing to S4 break detection and structured output.

[0128] During the iteration process, the system adopts an incremental update mechanism, performing OCR, layout analysis, semantic parsing, and dependency reconstruction processes only on newly added, modified, and deleted clauses. Unrevised clauses directly reuse historical structured data, and the condition dependency graph and rule chain network are updated locally in sync to improve update efficiency and achieve real-time monitoring and rapid iteration.

[0129] After iteration, standardized structured data and rule execution path information adapted to the latest version of the specification are generated, and the old and new data are replaced to ensure that the governance results are fully synchronized with the current civil aviation regulations and manuals. This can stably support practical applications such as the analysis of the impact of clause revisions and the tracing of conditional dependencies.

[0130] like Figure 3 As shown, this embodiment also provides a structured governance system for civil aviation regulations and manual data that integrates AI and OCR. The computer program running in this system executes the aforementioned structured governance method for civil aviation regulations and manual data that integrates AI and OCR, including: Data acquisition and preprocessing module 1 is used to preprocess the document images of civil aviation regulations and manuals, and extract the text data in the documents through OCR technology; Page Layout Analysis and Information Location Module 2: Based on text data, Page Layout Analysis and Information Location Module 2 identifies the page layout structure and divides the clause areas of the document to obtain structured clause units. Semantic parsing and rule network construction module 3 uses an artificial intelligence-based semantic parsing model to perform semantic parsing on structured clause units, constructing a semantic representation that includes entity information, conditional relationships, and obligation modalities; and applies semantic constraints to the semantic representation based on a civil aviation ontology library, constructing conditional dependencies between clauses, and then constructing a conditional dependency graph and a corresponding rule chain network based on the conditional dependencies, making the rule chain network computable and generating corresponding semantic evaluation information; The fracture detection and structured output module 4 performs logical consistency detection on the rule chain network based on the condition dependency graph, identifies the condition dependency fractures between clauses and the incomplete rule execution path; and completes or marks the fractured or abnormal rule chain by combining the semantic evaluation results generated by the semantic parsing process; then, it generates rule execution path information based on the condition dependency graph and outputs standardized structured data and rule execution path information. The data verification and dynamic update module 5 is used to verify and assess the quality of structured data and rule execution path information, and triggers iterative updates of the data acquisition and preprocessing module 1, layout analysis and information positioning module 2, semantic parsing and rule network construction module 3, and break detection and structured output module 4 based on the revision information of civil aviation regulations and manuals.

[0131] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A structured governance method for civil aviation regulations and manual data integrating AI and OCR, characterized in that, include: S1. Data Acquisition and Preprocessing: Preprocess the images of civil aviation regulations and manuals documents, and extract the text data from the documents using OCR technology; S2. Page Layout Analysis and Information Location: Based on text data, the document's page layout structure is identified and clause areas are divided to obtain structured clause units; S3. Semantic parsing and rule network construction: The structured clause unit is semantically parsed using an AI-based semantic parsing model to construct a semantic representation containing entity information, conditional relationships, and obligation modalities. The semantic representation is semantically constrained based on the civil aviation ontology library to construct conditional dependencies between clauses. Based on the conditional dependencies, a conditional dependency graph and the corresponding rule chain network are constructed to make the rule chain network computable and generate corresponding semantic evaluation information. S4. Break Detection and Structured Output: Based on the condition dependency graph, logical consistency detection is performed on the rule chain network to identify breaks in condition dependencies between clauses and incomplete rule execution paths; broken or abnormal rule chains are completed or marked based on the semantic evaluation results generated during the semantic parsing process; rule execution path information is generated based on the condition dependency graph; standardized structured data and rule execution path information are output, which are used to support the analysis of the propagation of the impact of clause revisions and the tracing of condition dependency paths; S5. Data Validation and Dynamic Update: Validate and assess the quality of structured data and rule execution path information, and trigger iterative updates of steps S1 to S4 based on revisions to civil aviation regulations and manuals.

2. The structured governance method for civil aviation regulations and manual data integrating AI and OCR as described in claim 1, characterized in that, The layout analysis and information location of S2 include the following steps: S2.1 Based on text data, use layout structure recognition technology to identify structural elements in the document and generate layout metadata containing position coordinates and hierarchical relationships; S2.

2. Use page layout metadata to divide the document into clause areas, cut continuous text into independent structured clause units, each structured clause unit shall contain at least a unique clause number, complete clause text content, spatial location information, and adjacency and reference relationship markers with other structured clause units in the same document. S2.

3. Perform boundary verification and error correction on the divided structured clause units. Through syntactic consistency detection and contextual coherence analysis, correct the regional overlap, missing or misaligned areas caused by layout structure recognition deviation.

3. The structured governance method for civil aviation regulations and manual data integrating AI and OCR as described in claim 1, characterized in that, In step S3, semantic parsing and semantic representation construction include the following steps: S3.

1. Use a lightweight semantic parsing model that has been fine-tuned with civil aviation regulations corpus, or a large language model guided by the civil aviation field's special prompt word project, as the main body for semantic parsing execution; S3.

2. Perform fine-grained semantic extraction on the complete clause text content of each structured clause unit. The extracted content includes entity information in the civil aviation field, conditional relationships corresponding to conditional predicates, and obligation modalities corresponding to statutory norms. S3.3 Based on the extraction results, construct a standardized "entity-condition-modality" triple semantic representation for each structured clause unit. The triple semantic representation is the smallest execution unit of semantic parsing.

4. The structured governance method for civil aviation regulations and manual data integrating AI and OCR as described in claim 3, characterized in that, In step S3, performing semantic constraints and constructing conditional dependencies includes the following steps: S3.

4. Using a civil aviation ontology library built based on CAAC civil aviation regulations and ICAO international civil aviation standards, semantic constraints and consistency checks are performed on the semantic representation of "entity-condition-modality" triples. S3.5 Based on the semantic representation of "entity-condition-modality" triples that complete semantic constraints, identify the logical relationships between different structured clause units and construct conditional dependencies between clauses; among which conditional dependencies include premise-conclusion type causal dependencies and reference-referenced type normative dependencies; S3.

6. Add a unique association identifier, dependency type tag, and confidence score to each set of conditional dependencies to complete the standardized construction of conditional dependencies.

5. The structured governance method for civil aviation regulations and manual data integrating AI and OCR as described in claim 4, characterized in that, In step S3, generating semantic evaluation information includes the following steps: S3.

7. Construct a condition dependency graph using individual structured clause units as the set of nodes and standardized condition dependencies as the set of directed edges. The core formal definition is: ； in, For conditional dependency graphs, A set of nodes that corresponds one-to-one with a structured clause unit with a unique clause number. This is the set of directed edges representing the conditional dependencies between corresponding clauses. Let be the set of edge weights corresponding to the confidence score of each conditional dependency, and The range of values ​​for the inner element is [0,1]. S3.8 Based on the standard syntax of description logic and rule engine, the semantic representation of the "entity-condition-modality" triples of each node in the condition dependency graph and the condition dependency relationship of each directed edge are formally encoded, transforming the structured clause unit into a machine-executable logical unit and the condition dependency relationship into a logical jump relationship between structured clause units. S3.

9. Based on the formally encoded logical units and logical jump relationships, construct a computable rule chain network. The specific verification process is as follows: In the condition dependency graph Extract simple directed paths with no repeating nodes as a single rule chain. All valid rule chains constitute a rule chain network; a single rule chain It covers two types of scenarios: complete logical paths and single-node independent paths; Calculate a single rule chain Completeness rating , The value range is [0,1]; based on A computability check was performed against a preset integrity threshold. Rule chains with a completeness threshold greater than or equal to the preset completeness threshold are considered valid rule chains, while those with a completeness threshold less than or equal to the preset completeness threshold are considered abnormal rule chains to be verified. The preset completeness threshold is a fixed configuration value within the closed interval [0,1]. S3.10, Node and Edge Attributes Based on Conditional Dependency Graphs The computability verification results are used to generate semantic evaluation information for the corresponding structured clause units.

6. The structured governance method for civil aviation regulations and manual data integrating AI and OCR as described in claim 5, characterized in that, In step S4, logical consistency detection and identification of conditional dependency breaks between clauses and incomplete rule execution paths are performed, including the following steps: S4.

1. Traverse all rule chains in the rule chain network and extract all abnormal rule chains to be verified as detection objects; S4.

2. Combining the node attributes, edge attributes, and semantic evaluation information of the conditional dependency graph, identify the conditional dependency break type for each chain of abnormal rules to be verified; the specific identification rules are as follows: Node-missing type break: The judgment rule is that the preceding structured clause unit in the rule chain is missing, resulting in the preceding condition dependency relationship of the subsequent structured clause unit having no matching subject; Edge failure type fracture: The judgment rule is that the confidence score of the condition dependency relationship between the structured clause units in the rule chain is lower than the preset failure threshold; Incomplete path breakage: The judgment rule is the integrity score of the rule chain. Below the preset integrity threshold; Semantic conflict type break: The determination rule is that there is a semantic conflict in the semantic representation of the "entity-condition-modality" triple of the preceding and following structured clause units in the rule chain.

7. The structured governance method for civil aviation regulations and manual data integrating AI and OCR as described in claim 6, characterized in that, In step S4, after identifying the condition-dependent break type, the rule chain is quantified and determined, including the following steps: S4.3 Construct a quantitative model for rule chain breakage and calculate the breakage score of a single rule chain. The calculation formula is: ； in, For the first The breakage score of the rule chain has a fixed value range of [0,1]. This represents the number of directed edges contained in the rule chain; S4.4 Set a preset fracture judgment threshold and score the fracture degree. Fracture determination is completed by comparing with the preset fracture determination threshold: when If the value is greater than or equal to the preset breakage judgment threshold, the rule chain is determined to be an abnormal rule chain with conditional breakage. when If the value is less than the preset breakage threshold, the rule chain is determined to be a valid rule chain without breakage. S4.5 For the abnormal rule chain that is determined to have condition-dependent fracture, associate it with the corresponding fracture type identified in S4.2, generate a fracture quantification judgment result containing fracture location, fracture type, and fracture degree score, and synchronously update the semantic evaluation information of the corresponding structured clause unit.

8. The structured governance method for civil aviation regulations and manual data integrating AI and OCR as described in claim 7, characterized in that, In step S4, broken or abnormal rule chains are completed or marked, and corresponding output information is generated, including the following steps: S4.6 Based on the fracture quantification judgment results, combined with semantic evaluation information and the civil aviation domain ontology library, semantic completion and repair are performed on the abnormal rule chains with conditional fractures. S4.7 For abnormal rule chains that cannot be semantically completed and repaired, mark the corresponding break type, break location and break degree score, form a fixed abnormal marker and include it in the semantic evaluation information. S4.

8. Based on the condition dependency graph after completion or marking, generate rule execution path information; S4.9 Output standardized structured data and rule execution path information. The rule execution path information is used to support the analysis of the impact of clause revisions and the tracing of conditional dependencies.

9. The structured governance method for civil aviation regulations and manual data integrating AI and OCR as described in claim 8, characterized in that, The data verification and dynamic update of S5 includes the following steps: S5.

1. Conduct a comprehensive verification and quality assessment of the standardized structured data and rule execution path information output by S4. The verification content includes data integrity, semantic consistency, and rule execution path coherence. The quality assessment results are associated with the semantic assessment information of the corresponding structured clause unit. S5.2 Real-time monitoring of revision information of civil aviation regulations and manuals. When revision actions are detected, trigger full-process iterative updates from S1 to S4, and synchronously update the condition dependency graph, rule chain network, fracture quantification judgment results and semantic evaluation information to ensure that standardized structured data and rule execution path information are consistent with the latest civil aviation regulations and manuals.

10. A structured governance system for civil aviation regulations and manual data integrating AI and OCR, characterized in that: The computer program of the structured governance system for civil aviation regulations and manuals data integrating AI and OCR is used to execute the structured governance method for civil aviation regulations and manuals data integrating AI and OCR as described in any one of claims 1-9, including: The data acquisition and preprocessing module (1) is used to preprocess the document images of civil aviation regulations and manuals, and extract the text data in the documents through OCR technology. Page layout analysis and information positioning module (2), the page layout analysis and information positioning module (2) based on text data, performs page structure recognition and clause area division on the document to obtain structured clause units; The semantic parsing and rule network construction module (3) performs semantic parsing on the structured clause units based on the semantic parsing model of artificial intelligence, and constructs a semantic representation containing entity information, conditional relationships and obligation modalities; and performs semantic constraints on the semantic representation based on the civil aviation domain ontology library, constructs conditional dependency relationships between clauses, and then constructs a conditional dependency graph and the corresponding rule chain network based on the conditional dependency relationships, so that the rule chain network has computability and generates corresponding semantic evaluation information; The fracture detection and structured output module (4) performs logical consistency detection on the rule chain network based on the condition dependency graph, identifies the condition dependency fracture between clauses and the incomplete rule execution path; and completes or marks the fractured or abnormal rule chain based on the semantic evaluation results generated by the semantic parsing process; and then generates rule execution path information based on the condition dependency graph, and outputs standardized structured data and rule execution path information. The data verification and dynamic update module (5) is used to verify and evaluate the structured data and rule execution path information, and trigger the iterative update of the data acquisition and preprocessing module (1), the layout analysis and information positioning module (2), the semantic parsing and rule network construction module (3), and the break detection and structured output module (4) based on the revision information of civil aviation regulations and manuals.

Images