Civil aviation transport total condition semantic model construction method, system and electronic equipment

Abstract

The disclosure provides a civil aviation transport general condition semantic model construction method and system, which uses a large language model to process unstructured civil aviation transport general condition text. Because the text is similar to the internal logical structure and expression method of legal norms, a prompt word can be constructed to guide the large language model to extract triples according to the legal structure specification, and a semantic knowledge graph is preliminarily constructed. Isolated nodes and / or sparse nodes with a node degree lower than a preset threshold are identified. Then, the large language model is used to further mine the civil aviation transport general condition text, and the potential relationship between the to-be-supplemented nodes and other entities is supplemented to obtain a completed semantic knowledge graph. Self-supervised logic refinement is performed to replace the semantically ambiguous relationship with accurate legal logic. All entities and relationships after refinement are integrated to generate a triple set and obtain a civil aviation transport general condition knowledge graph semantic model. Reliable data support is provided for civil aviation operation, service and compliance decision-making.

Description

Technical Field

[0001] This disclosure relates to the field of text processing technology for general conditions of civil aviation transportation, and more specifically, to a method for constructing a semantic model of general conditions of civil aviation transportation driven by a full-process large language model, a system for constructing a semantic model of general conditions of civil aviation transportation, an electronic device based on the method for constructing a semantic model of general conditions of civil aviation transportation, a computer-readable storage medium, and a computer program product. Background Technology

[0002] As a vital pillar of global economic and social development, the civil aviation industry exerts a broad and profound influence on modern society. With the continuous increase in global air traffic, civil aviation-related documents are constantly evolving in terms of complexity and diversity, exhibiting a similar expression and logical structure to laws and regulations, and thus possessing a high degree of rigor. To ensure efficient and reliable flight operations and improve service quality, airlines and civil aviation authorities are required to acquire and process massive amounts of NOTAMs, regulations, and standards in a timely and accurate manner.

[0003] With the rapid development of information technology, knowledge graphs have been widely applied in various fields. They can describe concepts, entities, and relationships in the objective world in a structured way, transforming both structured and unstructured data into intuitive and easy-to-understand graph structures, providing data support for downstream applications such as information retrieval and intelligent decision-making. In the civil aviation sector, with the continuous expansion of business and the increasing complexity of the regulatory system, the rapid construction of accurate and complete knowledge graphs is of great significance for improving civil aviation operational efficiency, ensuring operational safety and compliance, and enhancing service quality.

[0004] Currently, knowledge graph construction in the civil aviation field primarily relies on manual construction or traditional machine learning and natural language processing techniques. With the rapid development of generative large language models, their powerful contextual understanding capabilities have made it possible to apply them to knowledge graph construction. However, these methods have several shortcomings, such as a lack of logical understanding of legal and regulatory texts, chaotic text structure analysis, insufficient knowledge graph relationship analysis, and relationship generalization. In summary, existing knowledge graph construction methods in the civil aviation field are significantly deficient in terms of automation, accuracy, completeness, and relationship precision, severely hindering the application and development of knowledge graphs in the civil aviation sector. An innovative approach is urgently needed to address these issues. Summary of the Invention

[0005] The embodiments of this disclosure provide a method for constructing a semantic model of general conditions for civil aviation transportation driven by a full-process large language model, a system for constructing a semantic model of general conditions for civil aviation transportation, an electronic device, a computer-readable storage medium, and a computer program product based on the method for constructing a semantic model of general conditions for civil aviation transportation, which can at least partially solve the above-mentioned problems or other problems.

[0006] In a first aspect, the embodiments of this disclosure provide a semantic model construction method for the overall conditions of civil aviation transportation driven by a large language model throughout the entire process, including the following steps:

[0007] The initial construction steps involve using a natural language processing model to process the unstructured general conditions text for civil aviation transportation. Processing is carried out, among which This is the unstructured General Conditions of Civil Aviation Transport text, also known as the original text. It is constructed using prompt words. Guided by legal structural norms, the natural language processing model extracts triples from the text to initially construct a semantic knowledge graph. ,in The knowledge graph structure with nodes and relation edges, initially extracted by the large language model, is the initial knowledge graph;

[0008] The completion step involves analyzing the semantic knowledge graph constructed in the initial construction step to identify nodes that need to be supplemented. , The nodes to be supplemented are those not extracted during the initial extraction of the large model in the preliminary construction step. These nodes include: isolated nodes and / or nodes with a degree lower than a preset threshold. sparse nodes, The integer is a pre-set value representing the lower limit of the degree when a node is determined to be an isolated node; the natural language processing model is then used to delve deeper into the civil aviation transportation general condition text to mine and supplement the potential relationships between the node to be supplemented and other entities. The completed semantic knowledge graph is obtained, in which For nodes to be added and Potential relation edges between nodes;

[0009] The logical refinement step involves using the natural language processing model to perform self-supervised logical refinement of the relations in the completed semantic knowledge graph. This replaces semantically ambiguous relationships with precise legal logic, where... It is a self-refined logical function, primarily completed by the large language model;

[0010] The integration step integrates all entities and relations refined in the logical refinement step to generate the final set of triples, resulting in the semantic model of the Civil Aviation Transportation General Conditions Knowledge Graph. This is the final map.

[0011] In some embodiments of this disclosure, during the initial construction step, the natural language processing model, in accordance with legal structural norms, extracts data from the text. Extracting triples from the text specifically includes: extracting the text of the General Conditions of Civil Aviation Transport. The descriptions involving Condition, Result behavior, and Outcome are parsed and extracted into header entities in the triples. ,relation Tail-end entity in For the initial map The node part, For the initial map The relational edge part in the middle,

[0012] The triplet is expressed as follows:

[0013] in:

[0014] · Represents the prerequisite entity (Condition)

[0015] · Represents behavioral norms and relationships (Result)

[0016] · Represents the legal consequences (Outcome).

[0017] In some embodiments of this disclosure, the natural language processing model includes, in the initial construction step, a large language model. .

[0018] In some embodiments of this disclosure, the initial construction step further includes attribute nodeization to obtain independent entity nodes. , This refers to the part that, after attribute nodeization, is transformed from entity attributes into independent entity nodes: during the extraction of triples, the numbers appearing in the text are... and / or date Define a transformation function to convert entity attributes into independent entity nodes. (·),

[0019]

[0020] in, These represent the numerical attribute, date attribute, and weight attribute appearing in the general civil aviation transport document, respectively. Obtained from text using information extraction techniques, the type belongs to a predefined set. Discrete data units.

[0021] In some embodiments of this disclosure, prior to the completion step, a virtual mounting is included to initially construct the set of all isolated nodes in the semantic knowledge graph. Temporarily connect to a virtual node Used to temporarily connect all isolated nodes together, it is a node without any actual meaning. Note. and define the mount operator. For set Establish temporary association edges for each node in the process. :

[0022]

[0023] The resulting set of temporary edges is defined as follows: To assist in the isolation of the node The systematic identification and subsequent processing of [the data], including: ,in For mounting operators, used for orphaned nodes Each node in the network establishes a connection to the virtual node. Temporary associated edges, Let be the set of temporary edges formed by temporary associated edges established by the mounting operator.

[0024] In some embodiments of this disclosure, the preliminary construction step, after extracting triples, also includes relation deduplication, which removes duplicate triples where the head entity, relation, and tail entity are exactly the same by matching.

[0025] In some embodiments of this disclosure, the step of using the large language model to delve into the general conditions text of civil aviation transportation, and mining and supplementing the potential relationships between the node to be supplemented and other entities, includes: using the natural language processing model to execute a deep mining function. Delve deeper into the General Conditions of Civil Aviation Transportation text to identify and supplement the nodes that need to be added. Potential relationships with other entities are determined through iterative loops, with a threshold set at each iteration. Gradually relax, that is: > ,

[0026]

[0027]

[0028] in As a deep mining function, the large language model executes this function to delve deeper into the text and supplement the potential relationships between nodes. This is a function for counting node degrees. This is the original text. A preset threshold is used to determine whether a node is a sparse node. To perform the mining process iteratively in a loop, the first... The threshold is set in rounds, and this threshold will be gradually increased (i.e., ... > ).

[0029] In some embodiments of this disclosure, the completion step, after identifying the node to be completed and before mining and completing the potential relationships between the node to be completed and other entities, further includes: targeting these existing isolated nodes and sparse node sets. An iterative enhancement mechanism, called Graph-to-Text Loop, is introduced.

[0030] In some embodiments of this disclosure, the logic refinement step involves self-supervised logic refinement of the relationships in the completed semantic knowledge graph, replacing semantically ambiguous relationships with precise legal logic, including:

[0031] Constructed legal-grade predicate library = Based on the legal-grade predicate library, the fuzzy relations initially extracted are reaffirmed by the relation reiteration function. The remapping process is performed using the following formula:

[0032]

[0033] in, To build a legal-grade predicate library, To generate precise legal logic relationships based on the remapping of the legal-grade predicate library, This is a fragment of the original text, used to provide context when refining the logic. These are the semantically ambiguous relationships initially extracted.

[0034] In some embodiments of this disclosure, the integration step further includes defining a discriminant function. The constructed knowledge graph is post-processed using a natural language processing model, including at least relational logical direction correction, ensuring that the logical direction of the head entity pointing to the tail entity in the relation of a triple is correct. If the judgment result is If so, it means that the logical direction conforms to legal facts; if it is If so, it is determined to be a position swapping error during the extraction of the large model. This is a discriminant function used to post-process the constructed knowledge graph using a large language model, including checking the correctness of the logical direction of triples.

[0035] Secondly, embodiments of this disclosure provide a semantic model construction system for general conditions of civil aviation transportation, used to implement the aforementioned construction method, the system comprising:

[0036] The initial extraction module is configured to use a natural language processing model to extract unstructured civil aviation transport general conditions text. Processing is performed by constructing prompt words. Guided by legal structural norms, the natural language processing model extracts triples from the text to initially construct a semantic knowledge graph. ;

[0037] The graph completion module is configured to analyze the semantic knowledge graph built in the initial construction steps and identify nodes that need to be completed. The nodes to be supplemented include: isolated nodes and / or nodes with a degree lower than a preset threshold. Sparse nodes; for these existing isolated nodes and sparse node sets. A graph-to-text loop mechanism is introduced for iterative enhancement; the natural language processing model is then used to delve deeper into the civil aviation transportation general condition text to mine and supplement the potential relationships between the nodes to be supplemented and other entities. The completed semantic knowledge graph is obtained.

[0038] The relation refinement module is configured to perform self-supervised logical refinement of relations in the completed semantic knowledge graph using the natural language processing model. This replaces semantically ambiguous relationships with precise legal logic;

[0039] The integration module is configured to integrate all entities and relations refined through the logical refinement steps, generating the final set of triples to obtain the semantic model of the civil aviation transportation general conditions knowledge graph. .

[0040] Thirdly, this disclosure provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the semantic model construction method for general conditions of civil aviation transportation provided in any embodiment of the first aspect of this disclosure.

[0041] Another aspect of this disclosure provides a computer-readable storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it implements a semantic model construction method for general civil aviation transport conditions as provided in any embodiment of the first aspect of this disclosure.

[0042] In another aspect, this disclosure provides a computer program product storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the semantic model construction method for general civil aviation transport conditions provided in any embodiment of the first aspect of this disclosure.

[0043] According to at least one embodiment of the semantic model construction method, system, electronic device, computer-readable storage medium, and computer program product for the general conditions of civil aviation transport provided in this disclosure, a precise, complete, and logically rigorous semantic knowledge graph of the general conditions of civil aviation transport is systematically constructed through a process that integrates professional semantic modeling, automatic relation completion, and self-refinement. This method first utilizes natural language processing models (such as large language models) Based on the unique "condition-norm-consequence" logical structure of law, this method automatically extracts preliminary triples from unstructured text, fundamentally ensuring that the constructed knowledge accurately reflects the inherent professional logic of regulations and avoiding errors caused by traditional methods neglecting the legal reasoning structure. Addressing the issues of numerous isolated nodes and poor connectivity in the preliminary graph, this method innovatively introduces automatic identification and drives the model to mine potential relationships between isolated nodes and other entities in the graph, significantly improving relationship recall and the completeness and coherence of the graph, providing a comprehensive information network for integrated decision-making. To further enhance the reasoning usability of the graph, this method designs a self-supervised logical refinement based on a large model, automatically reviewing and refining the initially extracted fuzzy associations into words corresponding to legal logical relationships, achieving a qualitative change from surface associations to deep legal logic, enabling the graph to directly support complex reasoning such as liability determination and compliance review. Finally, after post-processing such as logical correction, a high-quality semantic model is generated for final use. This four-step process not only significantly improves construction efficiency through automation, but its high-precision, high-completeness knowledge graph provides reliable data support for civil aviation operations, services, and compliance decisions. The civil aviation knowledge graph encompasses diverse information such as professional terminology for civil aviation services and airline general conditions of transport, enabling efficient information integration and sharing, and providing technical support for the high-quality development of civil aviation services. For civil aviation staff, work guidance based on accurate knowledge graphs is more precise, reducing the risk of misjudgment when performing tasks and complying with regulations and standards. Airlines can also reduce operational risks and ensure safe and efficient flight operations by relying on accurate knowledge bases when formulating operational strategies and handling compliance matters. Attached Figure Description

[0044] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. Wherein:

[0045] Figure 1 This is a flowchart of a method for constructing a knowledge graph of general conditions for civil aviation transportation according to one embodiment of this disclosure;

[0046] Figure 2 This is a schematic diagram of some entities in step S1 of a method according to this disclosure;

[0047] Figure 3 This is a schematic diagram of a portion of the triples in step S1 of a method according to this disclosure;

[0048] Figure 4 This is a schematic diagram illustrating a portion of the civil aviation knowledge graph extracted in the embodiment.

[0049] Figure 5 This is a schematic diagram of the principle structure of the knowledge graph construction system for the general conditions of civil aviation transportation of the present invention.

[0050] Figure 6 The prompt words used in step S1 of the embodiment for extraction from the large model;

[0051] Figure 7 The prompt words used in step S3 of the embodiment for refining the relationships in the large model; and

[0052] Figure 8 This is a schematic diagram of the structure of a computer system suitable for implementing an electronic device according to one embodiment of the present disclosure;

[0053] Figure 9 This is a flowchart of a method for constructing a knowledge graph of general conditions for civil aviation transportation according to another embodiment of this disclosure. Detailed Implementation

[0054] To better understand this disclosure, various aspects of this disclosure will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are merely illustrative of exemplary embodiments of this disclosure and are not intended to limit the scope of this disclosure in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0055] It should be noted that in this specification, the terms "first," "second," "third," etc., are used only to distinguish one feature from another and do not imply any limitation on the features, especially not any order of precedence. Therefore, without departing from the teachings of this disclosure, the first predictive feature discussed herein may also be referred to as the second predictive feature, and vice versa.

[0056] As used herein, the terms “approximately,” “about,” and similar terms are used as terms indicating approximation, not as terms indicating degree, and are intended to describe inherent biases in measured or calculated values ​​that will be recognized by one of ordinary skill in the art.

[0057] It should also be understood that expressions such as "comprising," "including," "having," "containing," and / or "comprising" are open-ended rather than closed-ended expressions in this specification, indicating the presence of the stated features, elements, and / or components, but not excluding the presence of one or more other features, elements, components, and / or combinations thereof. Furthermore, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features, not just individual elements in the list. Additionally, when describing embodiments of this disclosure, the word "may" is used to mean "one or more embodiments of this disclosure." And the term "exemplary" is intended to refer to an example or illustration.

[0058] Unless otherwise specified, all terms used herein (including engineering and technical terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It should also be understood that, unless expressly stated in this disclosure, terms as defined in common dictionaries shall be interpreted as having the meaning consistent with their meaning in the context of the relevant art, and not as having an idealized or overly formalized meaning.

[0059] It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. Furthermore, unless explicitly limited or contradicted by the context, the specific steps included in the methods described in this disclosure are not limited to the order in which they are described, but can be performed in any order or in parallel. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0060] Furthermore, when the term "connection" or "linkage" is used in this disclosure, it may indicate direct or indirect contact between the corresponding components, unless otherwise expressly defined or deduced from the context.

[0061] The term “exemplary” as used herein means “serving as an example, implementation method, or illustration.” Any implementation method described herein as “exemplary” is not necessarily to be construed as superior to or better than other implementation methods.

[0062] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, systems, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0063] Currently, large language models are used in constructing knowledge graphs of texts related to civil aviation to improve operational efficiency and ensure operational safety and compliance. However, many problems still exist, such as:

[0064] The first problem involves leveraging the extensive coverage and strong contextual understanding capabilities of large language models to understand and process most texts, including civil aviation-related normative texts, and constructing a civil aviation knowledge graph. Because existing large language models are typically trained on general knowledge, traditional knowledge graph extraction methods often neglect the inherent reasoning structure of legal normative texts, extracting knowledge graphs based on fixed subject-verb-object formats. These texts possess similar expressions and logical structures to laws and regulations, exhibiting high rigor. This rigorous logical structure includes preconditions, normative content, and corresponding results. However, traditional methods fail to fully explore and utilize this structure, resulting in knowledge graphs that cannot accurately reflect their internal logical relationships. Ultimately, this leads to chaotic node content, unclear relationships, and limited effectiveness in practical applications.

[0065] The second problem is that the knowledge graph extracted based on the large language model has a very low ratio of relations to entities. That is, there are many nodes in the graph, but very few relation edges connecting these nodes. A large number of nodes are in an isolated state, the graph has poor connectivity, and there are a lot of isolated nodes. This is seriously inconsistent with the strong logical writing characteristics of general transportation conditions. Knowledge cannot form effective connections, which leads to reasoning obstacles, insufficient completeness and coherence, and reduces the practical value of the knowledge graph.

[0066] Thirdly, the knowledge graph relationships extracted using traditional methods are generalized and inaccurate. In the field of civil aviation compliance, different relational terms such as "basis," "trigger," and "restriction" have drastically different legal meanings. However, the relationships extracted by traditional methods are often too vague to accurately express these subtle differences. This limits the use of knowledge graphs to simple visualizations, preventing them from being used for in-depth logical reasoning and analysis, and thus failing to meet the actual needs of the civil aviation sector for precise knowledge graphs.

[0067] To address at least some or all of the aforementioned issues, and taking into full account the need for certain entities related to airport facilities in the civil aviation knowledge graph to establish effective relationships with relevant operational rules, safety standards, and other entities, it is essential to utilize the comprehensive and accurate information from the knowledge graph for analysis and decision-making.

[0068] When dealing with highly specialized fields like civil aviation, the large language model is trained and adjusted using legally binding texts to accurately grasp the complex logic and professional connotations contained within. This reduces comprehension biases and illusions. For example, when interpreting civil aviation safety regulations, the large model can accurately distinguish key elements such as preconditions, rights and obligations, and consequences between different regulatory clauses, without ignoring the unique logical structure of these laws and regulations, thus improving the accuracy and correctness of the knowledge graph extracted from them.

[0069] Furthermore, in the field of civil aviation compliance, different relational terms such as "basis," "trigger," and "restriction" have drastically different legal meanings. Improving the deep logical reasoning and analysis of knowledge graphs is crucial to meeting the actual needs of the civil aviation sector for accurate knowledge graphs. For example, in analyzing the determination of liability for civil aviation flight delays, enhancing the accuracy of relationships and accurately determining whether the delay was "triggered" by weather conditions or "restricted" by airport management regulations, based on the knowledge graph, is essential to ensuring the accuracy and efficiency of liability determination.

[0070] In order to at least solve some or all of the above problems, and taking into full account the characteristics and application scenarios of the above general conditions for civil aviation transportation, some embodiments of this disclosure provide a semantic model construction method for general conditions for civil aviation transportation driven by a full-process large language model. Figure 1 This is a flowchart of a semantic model construction method 1000 for general civil aviation transport conditions provided according to an exemplary embodiment of this disclosure.

[0071] like Figure 1 or Figure 9 As shown, the semantic model construction method 1000 for the general conditions of civil aviation transportation includes:

[0072] Step S1, utilize natural language processing models such as large language models For unstructured General Conditions of Civil Aviation Transport Processing is performed by constructing prompt words. Guiding large language models Following legal structural norms, triples are extracted from the text to initially construct a semantic knowledge graph. The triplet is expressed as follows:

[0073] in:

[0074] · This represents the prerequisite entity. (Condition)

[0075] · Represents behavioral norms and relationships (Result)

[0076] · Represents the legal consequences (Outcome);

[0077] Step S2 involves analyzing the semantic knowledge graph constructed in the initial construction step and identifying nodes that need to be supplemented. The nodes to be added include: isolated nodes and / or nodes with a degree lower than a preset threshold. The sparse nodes are then analyzed; the natural language processing model is used again to delve deeper into the civil aviation transportation general conditions text, mining and supplementing the potential relationships between the nodes to be supplemented and other entities. This yields a completed semantic knowledge graph; further, a large language model is used for deep mining functions. By thoroughly examining the General Conditions of Civil Aviation Transport, we can uncover and supplement the potential relationships between the nodes to be supplemented and other entities. ,in For node degree, This is the original text. This is a preset threshold. The execution method is iterative looping, with the threshold gradually widening in each iteration. > : Obtain the completed semantic knowledge graph.

[0078] Step S3, utilize natural language processing models such as large language models Self-supervised logical refinement of relations in the completed semantic knowledge graph. To resolve semantically ambiguous relationships Replace with precise legal logic; ;

[0079] Step S4 involves integrating all entities and relations refined through the logical refinement step to generate the final set of triples, thus obtaining the semantic model of the civil aviation transportation general conditions knowledge graph. .

[0080] Through a process that integrates professional semantic modeling, automatic relation completion, and self-refinement, a precise, complete, and logically rigorous semantic knowledge graph of the general conditions of civil aviation transportation is systematically constructed.

[0081] Since civil aviation-related texts have the same logical structure and expression as laws and regulations, a large language model is used to automatically extract preliminary triplets from unstructured texts based on the "condition-norm-consequence" logical structure unique to law. This fundamentally ensures that the constructed knowledge accurately reflects the inherent professional logic of the regulations and avoids errors caused by traditional methods that ignore the legal reasoning structure.

[0082] To address the issues of numerous isolated nodes and poor connectivity in the initial graph, this paper innovatively introduces a method to automatically identify and drive the model to mine potential relationships between isolated nodes and other entities in the graph. This significantly improves the relationship recall rate and the completeness and coherence of the graph, providing a comprehensive information network for integrated decision-making.

[0083] To further enhance the reasoning usability of the graph, this method designs a self-supervised logic refinement based on a large model, which automatically reviews and refines the initially extracted fuzzy associations into words corresponding to legal and logical relationships. This achieves a qualitative change from surface associations to deep legal and logical relationships, enabling the graph to directly support complex reasoning such as liability determination and compliance review.

[0084] After post-processing such as logical correction, a high-quality semantic model is generated for final use. This four-step process not only significantly improves the construction efficiency through automation, but also produces a high-precision, high-completeness knowledge graph that can provide reliable data support for civil aviation operations, services, and compliance decisions.

[0085] The constructed civil aviation knowledge graph encompasses diverse information on civil aviation service terminology and airline general conditions of transport, enabling efficient information integration and sharing, and providing technical support for the high-quality development of civil aviation services. For civil aviation personnel, work guidance based on an accurate knowledge graph is more precise, reducing the risk of misjudgment when performing tasks and complying with regulations and standards. Airlines can also reduce operational risks and ensure safe and efficient flight operations by relying on accurate knowledge bases when formulating operational strategies and handling compliance matters.

[0086] Step S1

[0087] Figure 9 This is a flowchart of a semantic model construction method 1000 for general civil aviation transport conditions provided according to an exemplary embodiment of this disclosure.

[0088] like Figure 9 As shown, in some embodiments of this disclosure, the semantic model construction method 1000 for general conditions of civil aviation transportation, S1 utilizes a large language model. For unstructured General Conditions of Civil Aviation Transport Processing is performed by constructing prompt words. Guided by legal structural norms, the large language model extracts triples from the text to initially construct a semantic knowledge graph. This involves using a large language model and writing a prompt to allow the model to perform preliminary triple extraction from the unstructured civil aviation transportation general condition text according to the Condition-Result-Outcome format. This step can be further refined as follows:

[0089] Step S11: Select the large language model.

[0090] We selected the large model for semi-automatic knowledge graph extraction by analyzing the ease of use and API costs of large language models both domestically and internationally.

[0091] For example, the first is the parsing accuracy of texts with Chinese legal semantic structure. This is the cornerstone of knowledge extraction quality. The focus is on examining the model's ability to parse long and complex sentences, passive voice, and technical terms (such as "offer" and "defense") in texts with Chinese legal characteristics, as well as its ability to accurately extract entities (such as law names and clause numbers), relationships (such as "applied to" and "unless otherwise provided"), and attributes (such as effective date and revision status) from unstructured text. Selection criteria can refer to the model's performance on Chinese legal benchmarking lists such as LawBench or CAIL, or conduct small-scale triplet extraction tests using typical legal provisions (such as Article XXX of the Civil Code) to compare the completeness and accuracy of the extraction results. The second is the ability to output structured data. Semi-automatic extraction requires the model to stably output a format that is easy for the program to parse. Prioritize large models that support JSON schema or function calling, and ensure they strictly adhere to a predefined schema (e.g., {"entity": "..., "relationship": "..., "attribute": ...}) through prompts to avoid parsing difficulties caused by free text. The selection criterion is whether the model can maintain consistent output format across multiple consecutive calls, with no missing fields or excessive content. Thirdly, consider the overall cost, including API call fees and engineering costs. Calculate the token price for both input (prompt) and output (completion), and estimate the total cost based on the expected volume of regulatory text to be processed. Also consider the stability of domestic access, API response latency, and whether high concurrency is supported. The selection criterion is to prioritize cost-effective models with good domestic accessibility within the budget, balancing the cost per call with processing speed. Fourthly, consider the context window length. Texts like General Conditions of Transport often contain numerous interconnected clauses. A sufficient context window (128K or more recommended) allows the model to process the entire transport conditions or related sections at once, thereby capturing complex relationships across clauses and avoiding omissions in relation extraction due to text truncation. The selection criterion is to ensure the window size covers the largest unit of text you process most frequently (such as a complete regulation). Fifthly, ease of use and ecosystem considerations include API documentation clarity, SDK completeness, community activity, and support for streaming output. For semi-automatic knowledge graph construction, it's also necessary to consider whether the model provider offers convenient data management tools or compatibility with commonly used development frameworks. The selection criteria are low learning curve, easy troubleshooting of technical issues, and rapid integration into your existing development workflow.

[0092] Large language model This example uses the domestic Deepseek large-scale model, which is accessed and interacted with via API calls. Two Deepseek large-scale models are used here: Deepseek-Chat and Deepseek-Reasoner. The Chat model is mainly used for shallow, general processing, while Deepseek-Reasoner is an inference model used to infer depth and hidden content. The model can be selected according to different needs. In this example, Deepseek-Reasoner is mainly used for processing at each stage.

[0093] S12, Select the unstructured general conditions text for civil aviation transport.

[0094] The semantic model construction method of the General Conditions of Civil Aviation Transport disclosed herein needs to be applicable to the general text of the General Conditions of Civil Aviation Transport.

[0095] The unstructured general conditions of civil aviation transport used in this example is the "China Eastern Airlines Domestic Transport General Conditions". The full text of this text is 25,846 words long and covers 21 aspects, including definitions, scope of application, tickets, fares and taxes. It has a wide range of coverage, standardized language descriptions, and a similar logical structure and narrative method to standard laws and regulations. It is a relatively standard unstructured text of clause type.

[0096] S13, by constructing prompt words to guide the large language model to extract triples from the text according to legal structural norms, a preliminary semantic knowledge graph is constructed. .

[0097] By editing the prompt, Deepseek-Reasoner was instructed to follow the legal structure and specifications outlined in the "China Eastern Airlines Domestic Carriage General Conditions" text. The Condition, Result, and Outcome (CRO) statements are parsed and extracted into triples, including the head entity within each triple. ,relation Combined tail entity, .in For the initial map The node part, For the initial map The relational edge part in the middle, Figure 6 An example of a prompt message is provided.

[0098] When editing the prompt, it's crucial to consider the unique characteristics of the General Conditions of Civil Aviation Transportation. The focus should shift beyond generic triple extraction to logically structured extraction: the model must not only "see" entities but also "understand" the internal constraints, responsibility allocation, dependencies, and exceptions of the General Conditions of Transportation. Only by forcing the separation of logical elements, refining relational terms, handling hierarchical references, and outputting structured data with logical labels can the practical needs of deep knowledge graph reasoning and accurate liability determination in the civil aviation compliance field be met.

[0099] The prompt requires mandatory separation of logical elements (Condition-Right-Outcome). The core of the General Conditions of Civil Aviation Transportation is a complete logical chain consisting of preconditions, rights and obligations, and legal consequences. The prompt must force the model to identify and separate these elements: preconditions (e.g., "when a flight is delayed due to weather," "if the carrier has fulfilled its obligation to inform"), actors and rights and obligations (e.g., "the carrier has the right," "passengers should," "the airport management authority must"), and results and exceptions (e.g., "exemption from liability," "not considered a violation," "compensation can be applied for"). Simultaneously, the output must explicitly connect the three elements through triples, for example, in the form of (Condition: weather reasons, triggers, Rule: delay handling) and (Rule: delay handling, results_in, Outcome: carrier exemption), ensuring the logical chain remains unbroken.

[0100] The prompt also requires refined definition and constraints on relational terms. In the general conditions of civil aviation transport, relational terms such as "based on," "triggered," "restricted," "exempted," and "refer to" have distinctly different legal effects. Therefore, the prompt should avoid vague terms such as "related" and "involved," and guide the model to select the most appropriate relation. Examples of relational term classifications can be provided: causal relationships (triggered_by, result_in, due_to), authorization and obligation relationships (has_right_to, is_obliged_to, is_prohibited_from), condition and exception relationships (subject_to, unless, regardless_of), and reference and basis relationships (based_on, in_accordance_with, refer_to). Simultaneously, a node can participate in multiple relations; for example, a flight delay may involve both "triggered" (weather) and "restricted" (air traffic control regulations) to achieve joint expression of multiple relations.

[0101] The prompt should explicitly handle the hierarchy and referencing structure of the General Conditions of Transport. Civil aviation General Conditions of Transport often contain chapters, sections, articles, clauses, items, and numerous mutual references. The prompt should pay attention to node naming to reflect the hierarchy; for example, the ID should be refined to [abbreviation of General Conditions of Transport] clause number: tag: name (e.g., CCAR121.123: Rule: Fuel Requirements). Simultaneously, it should forcibly extract references such as "in accordance with Article X" and "except as provided in Article Y," forming triples of (Rule: A, refers_to, Rule: B) to construct knowledge relationships between General Conditions of Transport. For "provisos" or exception clauses, the is_exception_to relationship should be separately marked to avoid logical confusion.

[0102] The Prompt should define an output format with deep logic. To support subsequent liability determination or compliance reasoning, the output should include logical type labels or modal tags to enhance the semantics of triples. For example, it should label whether a triple belongs to condition, rule, or outcome, and annotate the modality (mandatory must, prohibitive must_not, authoritative may). Sample formats should include logical type and modal information, such as {"from": "Condition: Weather Reason", "to": "Rule: Flight Delay Handling", "relation": "triggers", "logic_type": "condition"} and {"from": "Rule: Flight Delay Handling", "to": "Outcome: Carrier Exemption", "relation": "results_in", "modality": "permitted"}, ensuring that the output knowledge graph can carry deep logic.

[0103] The triplet is expressed as follows:

[0104] in:

[0105] · Represents the prerequisite entity (Condition)

[0106] · Represents behavioral norms and relationships (Result)

[0107] · Represents the legal consequences (Outcome).

[0108] S14 Attribute Nodeization During the process of extracting triples, the numbers appearing in the text are... and / or date Define a transformation function to convert entity attributes into independent entity nodes. (·);

[0109] .

[0110] Information extracted from text using information extraction techniques refers to data of a type belonging to a predefined set. Discrete data units. Among them, This indicates the date attribute that appears in the general conditions of civil aviation transport. "Number" represents the numerical attributes appearing in the general conditions of civil aviation transport, while "Weight" represents the weight attributes appearing in the general conditions of civil aviation transport.

[0111] By turning all numerical portions of the unstructured transport conditions text into nodes, these numerical portions, which would otherwise be node attributes and are relatively scarce, are transformed into independent nodes. The core purpose of turning time and numbers into "attribute nodes" is to transform these values, which carry key logical information in the transport conditions text, into independent entities in the knowledge graph that can actively participate in relationship building, rather than passive descriptive labels. This directly serves the overall goal of this technology: to build a knowledge graph that "accurately reflects its inherent professional logical relationships." This ensures that the final graph not only shows "what is there," but also clearly reveals "under what specific numerical conditions, what regulations are triggered, and what consequences result," thereby supporting in-depth compliance review, intelligent question answering, and decision analysis. For example, the same "Value: 72 hours" node may be associated with different entities such as "voluntary refund application deadline" and "latest time for disabled passengers to apply for special services." This significantly increases the connectivity of the knowledge graph, making graph-based relational queries and reasoning (e.g., querying all rules related to the "72-hour" time limit) more direct and efficient.

[0112] S15, relation deduplication, removes duplicate triples where the head entity, relation, and tail entity are exactly the same by matching.

[0113] In unstructured transport condition texts, language and structure are generally standardized, so there may be many identical triples. In this step, regularization matching is used to remove identical triples, that is, only one triple with the same head node, relation edge, and tail node is kept.

[0114] In this embodiment, the "China Eastern Airlines Domestic Transport General Conditions" are extracted, and preliminary merging of duplicate triples is performed to avoid duplication, resulting in the extraction of 702 entity nodes (hereinafter referred to as entities). Figure 2 The cut-off portion of the entity is shown. Figure 3 The image shows a portion of the 502 triplets.

[0115] S16, Initial construction of semantic knowledge graph .

[0116] Step S2

[0117] Step S21, virtual mounting, will initially construct all isolated nodes in the semantic knowledge graph. Temporarily connect to a virtual node Above, pay attention and define the mount operator. For set Establish temporary association edges for each node in the process. :

[0118]

[0119] The resulting set of temporary edges is defined as follows: To assist in the isolation of the node Systematic identification and subsequent processing of [the data].

[0120] in: This is to assist in the systematic identification and subsequent processing of isolated nodes.

[0121] Connecting all isolated nodes to a single, meaningless node facilitates the subsequent processing of these isolated nodes.

[0122] S22, Analyze the semantic knowledge graph constructed in the preliminary construction step of S16, and identify the nodes that need to be supplemented. The nodes to be supplemented include: isolated nodes and / or nodes with a degree lower than a preset threshold. sparse nodes;

[0123] S23, for these existing sets of isolated nodes and diluted nodes. We introduce a graph-to-text loop mechanism for iterative enhancement.

[0124] Step S23 involves performing loop closure processing on isolated nodes. The system identifies isolated nodes and sparse nodes in the graph, using these as queries. The large model is then used to further extract and complete the potential relationships between entities in the original text, finding any missing or deep-level relationships.

[0125] The main issue is that knowledge graphs constructed using traditional methods (including preliminary extraction based on large language models) suffer from "a large number of isolated nodes," resulting in "poor graph connectivity." Preliminary extraction (step S1) is limited by the coverage of a single prompt and the model's single-pass understanding depth of complex text, inevitably missing many potential relationships between entities. This leads to a "fragmented" graph, with many entities (nodes) lacking connections (edges). This fragmentation of knowledge prevents the formation of effective reasoning chains, severely weakening its application value. "Loop closure processing" is designed to systematically diagnose and repair this structural defect.

[0126] The graph-text loop, using identified isolated / sparse nodes as clues (Query), drives the large language model to re-examine the original transportation condition text in greater depth. This is equivalent to performing a targeted deep information retrieval. Instead of indiscriminately extracting all relationships, it focuses on answering: "In the original text, is there a semantic or logical relationship between this 'isolated' node (e.g., 'small animals') and other existing nodes in the graph (e.g., 'baggage transportation regulations', 'special passenger services') that was not discovered in the initial extraction, or that is more hidden?" This significantly improves the recall rate of relationship extraction.

[0127] S24, Utilize the large language model to delve into the civil aviation transportation general conditions text, and mine and supplement the potential relationships between the nodes to be supplemented and other entities. This yields the completed semantic knowledge graph.

[0128] Execute deep mining functions using the natural language processing model. Delve deeper into the General Conditions of Civil Aviation Transportation text to identify and supplement the nodes that need to be added. Potential relationships with other entities. It's important to note that the process of mining and supplementing potential relationships between the node to be supplemented and other entities is performed iteratively, with the threshold gradually widening in each iteration. > ,

[0129]

[0130]

[0131] in For node degree, This is the original text. The preset threshold;

[0132] The initial iteration sets the strictest threshold (e.g., degree = 0), focusing on using a large model to uncover the potential relationships of the most obvious "isolated nodes," thus solving the most prominent structural problems with maximum efficiency. After the first round of completion, previously isolated nodes may be connected, but some "sparse nodes" (e.g., degree = 1) may still have incomplete knowledge associations due to their sparse connections. At this point, the threshold is gradually relaxed (e.g., from degree = 0 to degree < 2), meaning the system expands the next round of mining to "nodes with fewer than 2 connections." This ensures that the completion work prioritizes solving the most critical problems and dynamically adjusts the focus as the graph improves, thereby pursuing optimal overall processing efficiency while ensuring the quality of the final graph.

[0133] In this embodiment, because this step requires deeper relation reasoning, Deepseek-Reasoner and a large language model with better reasoning performance are preferentially used, effectively reducing the problem of hidden relations not being extracted. After loop closure processing to complete the relations, the original 502 triples are completed into 701 triples. This step can be executed cyclically, gradually relaxing the conditions for loop closure processing, for example, from isolated nodes attached to virtual nodes to nodes with a degree less than 2.

[0134] Step S3

[0135] In step S3, a large language model is used to perform self-supervised logical refinement of the relations in the completed semantic knowledge graph. This paper proposes a method to replace semantically ambiguous relationships with precise legal logic, allowing the large model to re-examine all ambiguous relationships and refine them into legal-grade predicates (e.g., replacing related_to with part_of). A self-supervised logic refinement mechanism based on LLM is also proposed. A secondary deduction of the initial relationships achieves a qualitative leap from 'fuzzy semantic associations' to 'precise legal logic'.

[0136] Constructed legal-grade predicate library = Based on the legal-grade predicate library, the fuzzy relations initially extracted are reaffirmed by the relation reiteration function. The remapping process is performed using the following formula:

[0137] in This is a fragment from the original text.

[0138] This step ensures that the extracted relationships are legally rigorous and can support complex logical reasoning such as compliance reviews.

[0139] Because the relationships extracted using a large model in this example may be too vague or imprecise, this step also uses the Deepseek-Reasoner model to write prompt words (see details). Figure 7(Screenshot of the prompt) For all triples, the correctness and detail of the relationships are checked by comparison with the original text. Since texts like the General Conditions of Transport are rigorous, the relationships of the extracted triples must be detailed and accurate, not vague "there is a relationship". A large language model is used for self-checking, replacing vague predicates with precise legal-grade predicates.

[0140] The execution method can be as follows: By designing a concise prompt containing an expert-level predicate list, the large model is driven to perform self-examination and logical calibration on the initially extracted triples. Specifically, the source entity, target entity, and original fuzzy relation (such as "related") from each initial triple are filled into a preset prompt template, while the original General Conditions of Transport text corresponding to that relation is provided as context. Based on civil aviation legal logic, the model selects the best match from seven predefined precise predicates (such as TRIGGERS, CONSTRAINTS, ACCORDING_TO, etc.) and outputs only the English label. For example, the initial relation "(flight delay, related_to, weather reason)" is reclassified as TRIGGERS after the model compares it with the original text "delayed due to weather," thus achieving a qualitative change from fuzzy semantic association to precise legal relation. The entire process is iteratively executed on all triples to ensure that the final relation meets the rigor requirements of the General Conditions of Transport.

[0141] The execution method can also be as follows: First, a dedicated "relation refinement" prompt can be constructed. After the initial triple extraction is completed, an independent refinement prompt is designed, requiring the model to re-examine the extracted fuzzy relations (such as related_to, associated_with). The prompt includes: a legal rule base, context implantation, and a forced inference instruction. The legal rule base provides a list of allowed precise relation words in the General Conditions of Civil Aviation Transportation (such as triggers, complies_with, is_exempted_by), along with definitions and examples. Context implantation feeds the original General Conditions of Transportation text and the initially extracted triples to the model simultaneously, requiring it to determine the accuracy of the relations based on the original context. The mandatory deduction instruction explicitly requires the model to answer, "Based on the legislative intent of the clause, is the relationship between A and B a triggering, restrictive, or causal relationship?" Then, a self-supervised logical deduction (double verification) is performed. The model performs internal deduction through a "thinking chain," such as: locating elements, identifying conditional adverbs (e.g., "when..."), authorizing verbs (e.g., "has the right"), and result conjunctions (e.g., "therefore") in the text; matching the relational database, comparing the identified elements with a predefined relational database, for example, finding that "due to weather reasons" corresponds to "triggered_by" and "must not exceed" corresponds to "is_limited_by"; and eliminating ambiguity, if the same conjunction has multiple possibilities (e.g., "according to" can mean both causal and method), the model needs to determine its legal attribute based on the context.

[0142] Step S4

[0143] Step S41 integrates all entities and relations refined by the logical refinement step to generate the final set of triples, thus obtaining the semantic model of the civil aviation transportation general condition knowledge graph, which integrates all entities and relations.

[0144] Step S42, define the discriminant function. The constructed knowledge graph is post-processed using a natural language processing model, including at least relational logical direction correction, ensuring that the logical direction of the head entity pointing to the tail entity in the relation of a triple is correct. If the judgment result is If so, it means that the logical direction conforms to legal facts; if it is If so, it is determined that there was a position swapping error during the extraction of the large model.

[0145] Post-processing, including logical direction re-correction, and visualization, yields the final knowledge graph of civil aviation transportation conditions. Figure 4 A partial screenshot of the graph is shown.

[0146] It should be noted that this step is mainly based on the specific problems found and they are addressed. If there are no problems, a set of triples will be generated directly after step S3. Integration means integrating entities and relations to directly generate a set of triples.

[0147] If problems are found, such as the extracted triples containing a mix of Chinese and English, or unclear points of the leading and trailing entities in the logical relationships, post-processing can be performed. All relationships can be reviewed through the large model to check for any issues with their relationships. If problems are found, logical corrections can be made.

[0148] In other words, this step is not fixed in this example. The modified triples in S3 are manually reviewed, and some problems are found, including mixed Chinese and English in the relation edges and incorrect head and tail entity pointers in the triple relation edges. These specific problems are solved through additional post-processing steps, including using a mapping table to unify Chinese and English relations and using a large model to re-examine the work, focusing on the correctness of the logical direction.

[0149] Because large language models may be unstable in their output, or the original prompts may not be strictly regulated by language, the relation predicates in the graph may contain a mixture of English and Chinese (e.g., both "triggers" and "triggering", "part_of" and "belong to" may exist). To address this issue, a standardized mapping table (e.g., {"triggers": "triggering"}) can be constructed in the post-processing stage. A script can then automatically traverse all triples, match their "relation" fields with the mapping table, and replace them with standardized terms, thereby achieving terminology unification.

[0150] The positions of the "head entity" and "tail entity" in a triple may be reversed, leading to incorrect logical direction. For example, the correct logic should be "flight delay triggers passenger compensation," but it might be incorrectly extracted as "passenger compensation triggers flight delay." To address this, a new, concise prompt instruction model can be designed. Given a triple and its original context, determine whether its logical direction (A->B or B->A) is correct; if incorrect, output the corrected triple. Alternatively, triples that are questionable or potentially problematic (e.g., those with explicit directional relations like causeBy or leadsTo) along with their context can be submitted to a large language model (such as Deepseek-Reasoner) for review. Based on its understanding of causal and conditional logic, the model outputs the correctly oriented triple.

[0151] The method disclosed herein solves the problems of high difficulty, long cycle, and low degree of automation in the traditional manual and machine learning and deep learning knowledge graph construction process by using a large language model to build a semi-automated knowledge graph of civil aviation transportation conditions.

[0152] This publicly available method innovatively proposes a CRO semantic modeling framework for civil aviation compliance. Closely integrating the characteristics of texts like the General Conditions of Civil Aviation Transport (GCT), which share similar structural expressions with laws and regulations, it forces the large model to extract triples according to rules based on the three-part structure of legal logic (condition-norm-consequence) using prompt words, ensuring that each part forms a legitimate reasoning chain. This allows the knowledge extracted from the GCT text to accurately reflect its inherent professional logical relationships, avoiding knowledge graph errors caused by insufficient understanding of the professional domain by the large model and by traditional methods neglecting the reasoning structure of legal norms.

[0153] This disclosed method introduces a "Graph-to-Text Loop" mechanism to process isolated nodes generated after the initial extraction by the large model. The system automatically identifies isolated nodes and incorporates them into query statements to mine potential relationships by delving deeper into the original text using the large model. This approach effectively improves relationship recall, significantly reduces the number of isolated nodes in the knowledge graph, and establishes closer and more comprehensive connections between entities in the knowledge graph. The complete and coherent knowledge graph provides rich and comprehensive information support for various civil aviation applications. In flight scheduling scenarios, dispatchers can comprehensively consider more relevant factors to formulate more reasonable and scientific flight plans, thereby improving flight operation efficiency.

[0154] This publicly available method employs a large-scale model self-supervised relationship refinement mechanism, allowing the large model to act as a reviewer. It re-examines and refines fuzzy relationships into legally-grade predicates, achieving a shift from "fuzzy semantic associations" to "precise legal logic." In the field of civil aviation compliance, precise relationships are crucial for accurate logical reasoning. For example, in handling complex issues such as determining liability for flight delays and enforcing safety regulations, precise "trigger" and "restriction" relationships enable airlines and relevant management departments to quickly and accurately determine liability attribution and enforcement standards based on knowledge graphs. This avoids decision-making errors caused by fuzzy relationships, improves dispute resolution efficiency, and protects the legitimate rights and interests of all parties.

[0155] The semi-automatic knowledge graph extraction process proposed in this disclosure demonstrates excellent versatility and can be widely applied to all civil aviation regulatory documents, including laws and regulations. The civil aviation legal system is complex and diverse, covering multiple aspects such as flight safety, airport operations, passenger services, air transport, and domestic and international matters. Different types of regulatory documents differ significantly in format, content focus, and language. However, this process, with its scientifically designed prompts, can accurately adapt to the characteristics of various civil aviation legal texts. The civil aviation knowledge graph encompasses diverse information such as civil aviation service terminology and airline transport conditions, enabling efficient information integration and sharing, and providing technical support for the high-quality development of civil aviation services. For civil aviation personnel, work guidance based on accurate knowledge graphs is more precise, reducing the risk of misjudgment when performing tasks and complying with regulations and standards. Airlines can also reduce operational risks and ensure safe and efficient flight operations by relying on accurate knowledge bases when formulating operational strategies and handling compliance matters.

[0156] Some embodiments of this disclosure provide a semantic model construction system for general conditions of civil aviation transportation. Figure 5 This is a block diagram of a semantic model construction system 2000 for general civil aviation transport conditions provided according to an exemplary embodiment of this disclosure.

[0157] The system includes:

[0158] The initial extraction module is configured to process the unstructured civil aviation transport general conditions text using a large language model. By constructing prompt words, the large language model is guided to extract triples from the text according to the legal structure and norms, and a semantic knowledge graph is initially constructed.

[0159] The graph completion module is configured to analyze the semantic knowledge graph constructed in the initial construction steps, identify nodes to be completed, including isolated nodes and / or sparse nodes with a degree lower than a preset threshold; then, the large language model is used to delve into the general conditions text of civil aviation transportation to mine and complete the potential relationships between the nodes to be completed and other entities, resulting in the completed semantic knowledge graph.

[0160] The relation refinement module is configured to use a large language model to perform self-supervised logical refinement of relations in the completed semantic knowledge graph, replacing semantically ambiguous relations with precise legal logic.

[0161] The post-processing and visualization module is configured to integrate and post-process all entities and relationships refined through the logical refinement steps, generating the final set of triples to obtain the semantic model of the civil aviation transportation general condition knowledge graph.

[0162] In some publicly available embodiments, the preliminary extraction module is further configured to parse and extract the descriptions of Condition, Result behavior, and Outcome consequences in the Civil Aviation Transportation General Conditions text into head entity, relation, and tail entity in triples, respectively. Specifically, each step in step S1 above is executed to achieve the corresponding function.

[0163] In some publicly available embodiments, the preliminary extraction module is also configured to perform attribute nodeization: during the extraction of triples, numbers and / or dates appearing in the text are converted from entity attributes into independent entity nodes. Specifically, each step in step S1 above is executed to achieve the corresponding function.

[0164] In some publicly available embodiments, the initial extraction module is also configured to perform relation deduplication, removing duplicate triples where the head entity, relation, and tail entity are identical through matching. Specifically, each step in step S1 above is executed to achieve the corresponding function.

[0165] In some publicly available embodiments, the graph completion module is further configured to perform virtual mounting before the completion step, temporarily connecting all isolated nodes in the initially constructed semantic knowledge graph to a virtual node to assist in the systematic identification and subsequent processing of isolated nodes. Specifically, each step in step S2 above is executed to achieve the corresponding function.

[0166] In some publicly available embodiments, the graph completion module is further configured to perform the mining and supplementation of potential relationships between nodes to be supplemented and other entities in a loop-iterative manner, with the preset threshold gradually relaxed in each iteration, specifically executing each step in step S2 above to achieve the corresponding function.

[0167] In some embodiments, the integrated model is further configured with post-processing, including at least logical direction correction, ensuring that the logical direction of the head entity pointing to the tail entity in the triplet relationship is correct. Specifically, each step in step S4 above is executed to achieve the corresponding function.

[0168] Figure 8 This is a schematic diagram of the structure of a computer system suitable for implementing an electronic device according to one embodiment of the present disclosure.

[0169] refer to Figure 8 The diagram illustrates a structural schematic of an electronic device 400 suitable for implementing embodiments of the present disclosure. Terminal devices in embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 8The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0170] like Figure 8 As shown, electronic device 400 may include a processing system (e.g., central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from storage system 408 into random access memory (RAM) 403. RAM 403 also stores various programs and data required for the operation of electronic device 400. Processing system 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0171] Typically, the following systems can be connected to I / O interface 405: input systems 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output systems 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage systems 408 including, for example, magnetic tapes, hard disks, etc.; and communication systems 609. Communication system 609 allows electronic device 400 to communicate wirelessly or wiredly with other devices to exchange data. Although... Figure 8 Electronic devices 400 with various systems are shown, but it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented or have alternatively. Figure 5 Each box shown can represent a system, or multiple systems as needed.

[0172] Furthermore, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication system 609, or installed from a storage system 408, or installed from a ROM 402. When the computer program is executed by a processing system 401, it performs the functions defined in the methods of embodiments of this disclosure. It should be noted that the computer-readable medium in embodiments of this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, system, or device. In embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, system, or device. Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0173] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: process the unstructured Civil Aviation Transportation General Conditions text using a large language model; guide the large language model to extract triples from the text according to legal structural norms by constructing prompt words, and initially construct a semantic knowledge graph; analyze the semantic knowledge graph constructed in the initial construction step, identifying nodes to be supplemented, including: isolated nodes and / or sparse nodes with a degree below a preset threshold; further utilize the large language model to delve into the Civil Aviation Transportation General Conditions text, mining and supplementing the potential relationships between the nodes to be supplemented and other entities, obtaining a completed semantic knowledge graph; use the large language model to perform self-supervised logical refinement of the relationships in the completed semantic knowledge graph, replacing semantically ambiguous relationships with precise legal logic; integrate all entities and relationships refined by the logical refinement step to generate the final set of triples, obtaining the semantic model of the Civil Aviation Transportation General Conditions knowledge graph. Computer program code for performing the operations of embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0174] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0175] The units described in the embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an observation unit, an estimation unit, and a positioning unit. The names of these units do not necessarily limit the specific unit; for example, a positioning unit may also be described as a "positioning object location information unit."

[0176] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for constructing a semantic model of general conditions for civil aviation transportation, characterized in that, The method includes the following steps: The initial construction steps involve using a natural language processing model to process the unstructured general conditions text for civil aviation transportation. Processing is performed by constructing prompt words. Guided by legal structural norms, the natural language processing model extracts triples from the text to initially construct a semantic knowledge graph. ; The completion step involves analyzing the semantic knowledge graph constructed in the initial construction step to identify nodes that need to be supplemented. The nodes to be supplemented include: isolated nodes and / or nodes with a degree lower than a preset threshold. The sparse nodes; the natural language processing model is then used to delve deeper into the civil aviation transportation general conditions text to mine and supplement the nodes to be supplemented. Potential relationships between nodes The completed semantic knowledge graph is obtained. The logical refinement step involves using the natural language processing model to perform self-supervised logical refinement of the relations in the completed semantic knowledge graph. This replaces semantically ambiguous relationships with precise legal logic; The integration step integrates all entities and relations refined in the logical refinement step to generate the final set of triples, resulting in the semantic model of the Civil Aviation Transportation General Conditions Knowledge Graph. .

2. The semantic model construction method for the general conditions of civil aviation transportation according to claim 1, characterized in that, In the initial construction step, the natural language processing model, in accordance with legal structural norms, extracts data from the text. Extracting triples from the text specifically includes: extracting the text of the General Conditions of Civil Aviation Transport. The descriptions involving Condition, Result behavior, and Outcome are parsed and extracted into header entities in the triples. ,relation Tail-end entity ,in For the initial map The node part, For the initial map The relational edge part in the middle, The triplet is expressed as follows: in: · Represents the prerequisite entity (Condition). · Represents behavioral norms and relationships (Result). · Represents the legal consequences (Outcome); And / or, in the initial construction step, the natural language processing model includes: a large language model .

3. The semantic model construction method for the general conditions of civil aviation transportation according to claim 1, characterized in that, The initial construction steps also include attribute nodeization to obtain independent entity nodes. During the triple extraction process, the numbers appearing in the text are... and / or date Define a transformation function to convert entity attributes into independent entity nodes. (·), ， in, These represent the numerical attribute, date attribute, and weight attribute appearing in the general civil aviation transport document, respectively. Obtained from text using information extraction techniques, the type belongs to a predefined set. Discrete data units.

4. The semantic model construction method for the general conditions of civil aviation transportation according to claim 1, characterized in that, Before the completion step, virtual mounting is also included, which will initially construct the set of all isolated nodes in the semantic knowledge graph. Temporarily connect to a virtual node ,Notice and define the mount operator. For set Establish temporary association edges for each node in the process. : ， The resulting set of temporary edges is defined as follows: To assist in the isolation of the node The systematic identification and subsequent processing of [the data], including: .

5. The semantic model of the general conditions of civil aviation transport according to claim 1 The construction method is characterized by, The completion step involves using the large language model to delve into the general conditions text of civil aviation transportation, mining and supplementing the nodes to be supplemented. The potential relationships between nodes include: performing deep mining functions using the natural language processing model. Delve deeper into the General Conditions of Civil Aviation Transportation text to identify and supplement the nodes that need to be added. and The potential relationships between nodes are determined through iterative loops, with a threshold value set at the end of each iteration. Gradually relax, that is: > , ， ， in As a deep mining function, the large language model executes this function to delve deeper into the text and supplement the potential relationships between nodes. This is a function for counting node degrees. This is the original text. A preset threshold is used to determine whether a node is a sparse node. To perform the mining process iteratively in a loop, the first... The threshold is set in rounds, and this threshold will be gradually increased (i.e., ... > ); And / or, The completion step, after identifying the node to be completed and before mining and completing the potential relationships between the node to be completed and other entities, also includes: addressing these existing isolated nodes and sparse node sets. An iterative enhancement mechanism, called Graph-to-Text Loop, is introduced.

6. The semantic model of the general conditions of civil aviation transport according to claim 1 The construction method is characterized by, In the logical refinement step, the relationships in the completed semantic knowledge graph are subjected to self-supervised logical refinement, replacing semantically ambiguous relationships with precise legal logic, including: The legal-grade predicate library constructed = Based on the legal-grade predicate library, the fuzzy relations initially extracted are reaffirmed by the relation reiteration function. The remapping process is performed using the following formula: ， in, To build a legal-grade predicate library, To generate precise legal logic relationships based on the remapping of the legal-grade predicate library, This is a fragment of the original text, used to provide context when refining the logic. These are the semantically ambiguous relationships extracted initially.

7. The semantic model of the general conditions of civil aviation transport according to claim 1 The construction method is characterized by, The integration process also includes defining a discriminant function. The constructed knowledge graph is post-processed using a natural language processing model, including at least relational logical direction correction, ensuring that the logical direction of the head entity pointing to the tail entity in the relation of a triple is correct. If the judgment result is If so, it means that the logical direction conforms to legal facts; if it is If so, it is determined to be a position swapping error during the extraction of the large model. This is a discriminant function used to post-process the constructed knowledge graph using a large language model, including checking the correctness of the logical direction of triples.

8. A semantic model construction system for general conditions of civil aviation transportation, used to implement the method as described in any one of claims 1-7, characterized in that, The system includes: The initial extraction module is configured to use a natural language processing model to extract unstructured civil aviation transport general conditions text. Processing is performed by constructing prompt words. Guided by legal structural norms, the natural language processing model extracts triples from the text to initially construct a semantic knowledge graph. ; The graph completion module is configured to analyze the semantic knowledge graph built in the initial construction steps and identify nodes that need to be completed. The nodes to be supplemented include: isolated nodes and / or nodes with a degree lower than a preset threshold. Sparse nodes; for these existing isolated nodes and sparse node sets. A graph-to-text loop mechanism is introduced for iterative enhancement; the natural language processing model is then used to delve deeper into the civil aviation transportation general condition text to mine and supplement the potential relationships between the nodes to be supplemented and other entities. The completed semantic knowledge graph is obtained. The relation refinement module is configured to perform self-supervised logical refinement of relations in the completed semantic knowledge graph using the natural language processing model. This replaces semantically ambiguous relationships with precise legal logic; The integration module is configured to integrate all entities and relations refined through the logical refinement steps, generating the final set of triples to obtain the semantic model of the civil aviation transportation general conditions knowledge graph. .

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the semantic model construction method for the general conditions of civil aviation transportation as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: At least one processor; The memory is communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the semantic model construction method according to any one of claims 1-7.

Images