An intelligent structured parsing method for air travel change rules

Abstract

This invention relates to the field of natural language processing technology and discloses an intelligent structured parsing method for airline refund and change rules. The method includes: receiving a ticket refund and change fee query request sent by a client and obtaining the unstructured refund and change rule text and ticket context information corresponding to the request; inputting the unstructured refund and change rule text into a preset artificial intelligence parsing engine to extract semantic entities related to refunds and changes; performing matrix mapping on the semantic entities according to a preset multidimensional data model to generate a multidimensional rule matrix containing time windows and dynamic rates; performing logical confidence evaluation and cross-validation on the multidimensional rule matrix to obtain the validation result; if the validation result is successful, automatically calculating the price based on the multidimensional rule matrix, ticket context information, and current system time, and outputting a structured refund and change fee result. This solution effectively improves the accuracy, compliance, and interpretability of refund and change pricing.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and more specifically, to an intelligent structured parsing method for airline refund and rebooking rules. Background Technology

[0002] In current air passenger ticketing systems, airlines' refund and rebooking rules are usually in the form of unstructured, long text in natural language. These rules are logically complex, vary in expression, and often include multi-dimensional time windows, relative proportion calculations, and special conditions (such as no-show penalties for missed flights).

[0003] Existing processing methods primarily rely on manual reading and pricing, or simple rule matching based on regular expressions. However, manual processing is inefficient and error-prone, while regular expression matching struggles to cover the ever-changing sentence structures and implicit logic. Furthermore, directly applying large language models (LLMs) for pricing carries the risk of "illusion," potentially leading to output results that significantly deviate from actual business benchmarks and fail to meet the stringent accuracy requirements of high-risk financial settlement scenarios. Therefore, this paper proposes an intelligent structured parsing method for airline refund and rebooking rules. Summary of the Invention

[0004] In view of this, the present invention proposes an intelligent structured parsing method for airline refund and change rules, aiming to solve the technical problems in the existing technology, such as low efficiency due to reliance on manual processing of airline refund and change rules, difficulty in covering complex semantic logic by traditional automation methods, and the risk of inaccurate pricing due to the illusion of directly applying large models, which cannot be directly applied to financial settlement.

[0005] To achieve the above objectives, this invention provides an intelligent structured parsing method for airline refund and rebooking rules, comprising: Receive a ticket refund and change fee query request sent by the client, and obtain the unstructured refund and change rule text and ticket context information corresponding to the request; The unstructured refund and rescheduling rule text is input into a preset artificial intelligence parsing engine to extract semantic entities related to refunds and rescheduling; The semantic entities are mapped into a matrix according to a preset multidimensional data model to generate a multidimensional rule matrix containing time windows and dynamic rates. Perform logical confidence evaluation and cross-validation on the multidimensional rule matrix to obtain the validation results; If the verification result is successful, the price is automatically calculated based on the multidimensional rule matrix, the ticket context information, and the current system time, and a structured refund and change fee result is output.

[0006] Preferably, before inputting the unstructured refund / refund rule text into a preset artificial intelligence parsing engine, the method further includes: The unstructured refund / rescheduling rule text is preprocessed to remove irrelevant and redundant characters; Using a text segmentation algorithm, the preprocessed unstructured refund and rebooking rule text is divided into independent business logic segments to generate cleaned text blocks to be parsed. The business logic segments include at least a refund rule segment, a rebooking rule segment, and a missed flight supplementary rule segment.

[0007] Preferably, the unstructured cancellation and rescheduling rule text is input into a preset artificial intelligence parsing engine, specifically including: Identify the service type based on the ticket refund and change fee inquiry request; Based on the business type, target prompt words containing data extraction instructions and output specifications are dynamically assembled from a preset prompt word library; The target prompt, the cleaned text block, and the ticket context information are input into the artificial intelligence parsing engine.

[0008] Preferably, the semantic entities are mapped into a matrix according to a preset multidimensional data model to generate a multidimensional rule matrix containing time windows and dynamic rates, specifically including: Extract time descriptors from the semantic entities and determine the scheduled flight departure time as a unified time calculation benchmark anchor point; The extracted time descriptors are converted into a unified time unit. Based on the logical relationship between the transformed time descriptor and the time calculation benchmark anchor point, a continuous time interval vector with opening and closing attributes is constructed.

[0009] Preferably, generating a multidimensional rule matrix that includes time windows and dynamic rates further includes: For each continuous time interval vector, extract the corresponding rate data, separate and record its numerical part and currency code; The rate data is determined to be represented by either an absolute value calculation or a relative proportion calculation. If the representation type is relative proportion calculation, then the calculation base object corresponding to the relative proportion calculation is further extracted. The calculation base object includes the ticket price, the actual ticket price, or the fare for the chargeable flight segment.

[0010] Preferably, generating a multidimensional rule matrix that includes time windows and dynamic rates further includes: Independently extract the additional conditions for missed flights and their corresponding time limits from the semantic entities; The system determines the additional fee stacking logic when the missed flight additional conditions are triggered. The additional fee stacking logic includes replacement logic that replaces the original time window fee, or stacking logic that adds a fee on top of the original time window fee. The continuous time interval vector, the rate data, the computational base object, and the cost superposition logic of the missed flight additional conditions are used as independent dimensions to assemble and generate the computer-addressable multidimensional rule matrix.

[0011] Preferably, performing logical confidence evaluation and cross-validation on the multidimensional rule matrix specifically includes: Map all time windows in the multidimensional rule matrix onto a one-dimensional linear time axis; Calculate the boundary values ​​of adjacent time windows and verify whether there are time blind spots with undefined rates in the time intervals on the one-dimensional linear time axis; Verify whether the same point in time is simultaneously included in two different time windows with different rates to detect logical overlap.

[0012] Preferably, performing logical confidence evaluation and cross-validation on the multidimensional rule matrix further includes: Obtain the total face value from the ticket context information; If the rate data in the multidimensional rule matrix is ​​represented as an absolute value, then verify whether the absolute value is less than or equal to the total face value. If the rate data in the multidimensional rule matrix is ​​represented as a relative proportion, then it is verified whether the relative proportion value is strictly distributed within the preset legal percentage closed interval.

[0013] Preferably, after obtaining the verification result, the method further includes: Based on the pass status of each cross-validation, a preset weighting algorithm is used to calculate the logical confidence score for the multidimensional rule matrix. If the logical confidence score is lower than the preset confidence threshold, or if a business extreme value conflict is captured in the cross-validation, the automated pricing process will be truncated. Generate an exception data snapshot containing an error type identifier, and downgrade the request to a manual review queue.

[0014] Preferably, based on the multidimensional rule matrix, the ticket context information, and the current system time, automated pricing is performed to output structured refund and change fee results, specifically including: Calculate the precise time interval vector that the current system time matches in the multidimensional rule matrix; The rate rules corresponding to the precise time interval vector are matched with the calculation base object to calculate the price, and when there is a missed flight indicator, the additional fee is calculated together to generate a precise refund and rebooking fee value; The precise refund and change fee values ​​and details are converted into standard data interaction messages and sent to the downstream ticketing settlement system, and explanatory text is generated and fed back to the client.

[0015] This application discloses an intelligent structured parsing method for airline refund and rebooking rules. This method inputs unstructured refund and rebooking rule text into a preset artificial intelligence parsing engine to extract semantic entities. These entities are then matrix-mapped according to a preset multidimensional data model to generate a multidimensional rule matrix containing time windows and dynamic rates. This successfully transforms fuzzy natural language rules into a low-level data structure with strict mathematical logic that can be directly addressed and computed by a computer, fundamentally solving the problems of accuracy and reliability in rule parsing.

[0016] The solution incorporates key steps of logical confidence assessment and cross-validation of the multi-dimensional rule matrix. This step proactively detects and intercepts unreasonable results (such as time blind spots, logical overlaps, and fees exceeding ticket prices) caused by model parsing errors or contradictions in the original rules. Only when the validation result passes is subsequent automated pricing allowed. This closed-loop architecture of "parsing-validation-execution" effectively isolates the risks between non-deterministic AI output and deterministic financial settlement, ensuring the safety and reliability of the final output of refund and change fees.

[0017] This solution not only achieves millisecond-level automated pricing, greatly improving processing efficiency, but also establishes a robust fault-tolerance system through a confidence assessment mechanism. When verification fails or the confidence level is insufficient, the system can automatically generate an abnormal snapshot and degrade the request to a manual review queue, thus avoiding erroneous pricing, ensuring user experience, and forming a highly robust and complete business closed loop. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart illustrating an intelligent structured parsing method for airline refund and rebooking rules, provided as an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] like Figure 1 As shown in some embodiments of this application, this embodiment provides an intelligent structured parsing method for airline refund and rebooking rules. Specifically, the method includes the following steps: Step S101: Receive a ticket refund and change fee query request sent by the client, and obtain the unstructured refund and change rule text and ticket context information corresponding to the request.

[0021] The core of this step lies in the aggregation of all context data. Specifically, the execution device (such as a ticketing server or intelligent pricing engine node) first needs to listen for and receive refund or rescheduling fee inquiry requests initiated by user clients (such as airline websites, mobile apps, or third-party OTA platforms). Upon receiving the request, the system does not simply process an isolated instruction, but proactively and comprehensively retrieves two types of key information closely related to the request from multiple data sources in the background: unstructured refund and rescheduling rule text and ticket context information.

[0022] Unstructured refund and change policy texts refer to the original, long texts issued by airlines through industry-standard channels (such as the ATPCO fare system) that describe the specific refund and change policies for a particular ticket. This text is typically in natural language and contains complex conditions, time windows, and rate descriptions, without any structured processing.

[0023] Ticket context information is a set of metadata directly bound to a specific ticket instance, used to provide precise contextual anchors for subsequent rule parsing and fare calculation. This information typically includes, but is not limited to: a unique ticket identifier (such as ticket number), the ticket's fare base code, the specific ticketing time, the flight's scheduled departure time, the passenger type (adult / child, etc.), and the total face value of the ticket.

[0024] By aggregating these two types of information, the system provides the subsequent intelligent analysis engine with a complete puzzle to be solved and clues to its solution, ensuring the relevance and accuracy of the analysis process.

[0025] For example, suppose a passenger uses an airline's mobile app to initiate a "rebooking fee inquiry" request for a ticket with ticket number 123-123456789. The system receives the request and uses its built-in intent recognition module to determine that the transaction type is "rebooking".

[0026] The system retrieves a complete unstructured text segment from the fare rule database based on the ticket's basic fare code (e.g., YX14D). This text segment might contain the following: "This fare allows for rescheduling. Rescheduling 7 days or more before departure is free; a 10% fee will be charged for rescheduling between 72 hours and 7 days before departure; a 20% fee will be charged for rescheduling between 24 hours and 72 hours before departure; a 50% fee will be charged for rescheduling within 24 hours of departure or for missed flights." At the same time, the system extracts detailed contextual information of the ticket from the core ticketing system, such as: ticket issuance time is 10:00 on May 1, 2024, flight scheduled departure time is 14:00 on May 15, 2024, total ticket value is RMB 1,000, and passenger type is adult.

[0027] At this point, the system has successfully completed step S101, preparing all the necessary input data for the next step of rule parsing.

[0028] It should be noted that, in specific implementation scenarios, a more refined business type identification scheme can be adopted based on the above approach. That is, in addition to the basic binary classification of "refund" or "rebooking," the system can further identify more granular business scenarios, such as "voluntary refund," "involuntary refund," "same-class rebooking," or "upgrade rebooking." This refined identification can be used to more accurately construct dynamic prompts in subsequent steps, thereby guiding the parsing engine to focus on the parts of the rule text most relevant to the current scenario.

[0029] A dynamic extension scheme for contextual information is adopted, meaning that the ticket context information is not limited to the static data mentioned above, but can also dynamically include real-time status information related to the time of query. For example, the system can obtain and transmit whether the current ticket has been partially used (e.g., the outbound journey has been completed), whether there are special promotional activity tags, or the passenger's membership level information. This dynamic context can also serve as an important reference for the parsing engine to handle complex rules that depend on passenger status or external conditions. All of the above optional schemes are within the scope of protection of this application.

[0030] Step S102: Input the unstructured refund and rescheduling rule text into the preset artificial intelligence parsing engine to extract semantic entities related to refunds and rescheduling.

[0031] The core of this step lies in utilizing artificial intelligence technology for deep semantic understanding and information extraction. In this step, the system takes the raw, lengthy, and inconsistently formatted refund and rescheduling rule text obtained in step S101 as the main input and feeds it into a pre-trained and optimized artificial intelligence parsing engine. This engine is not a general-purpose large model, but a customized intelligent agent that integrates a domain knowledge base and specific prompt engineering strategies.

[0032] The core task of this engine is to perform Named Entity Recognition (NER) and Relation Extraction. It can accurately identify all semantic entities in the text that are directly related to the refund and rebooking business logic. These entities are the basic elements that constitute the subsequent multi-dimensional rule matrix. Typical semantic entities include, but are not limited to: time descriptors, such as "7 days before departure," "within 72 hours," "flight date," etc.; fare data, such as "10%," "500 yuan," "free," etc.; business actions, such as "refund," "rebooking," "missed flight," etc.; calculation base objects, such as "face value," "actual fare," "fuel surcharge," etc.; and special conditions, such as "missed flight (No-Show)," "involuntary change," etc.

[0033] Through this process, the engine extracts the chaotic text information into a set of clear, discrete, and structured data points (i.e., semantic entities) with explicit business meaning, laying a solid data foundation for the next step of matrix mapping.

[0034] For example, suppose the system inputs the unstructured rule text it obtains: "This fare allows for rebooking. Free for bookings made 7 days or more before departure; 10% fee for bookings made 72 hours to 7 days before departure; 20% fee for bookings made 24 hours to 72 hours before departure; 50% fee for bookings made within 24 hours of departure or if the flight is missed." into the AI ​​parsing engine.

[0035] After analysis, the engine will extract the following key semantic entities: For fares "7 days or more before departure": Time descriptor: >= 7 days; Rate data: 0%; Calculation base object: Fare; For "72 hours to 7 days (excluding) before departure": Time descriptor: [72 hours, 7 days); Rate data: 10%; Calculation base object: Fare; For "24 hours to 72 hours (excluding) before departure": Time descriptor: [24 hours, 72 hours); Rate data: 20%; Calculation base object: ticket price; For flights "within 24 hours (excluding) before departure or missed": Time descriptor: <24 hours; Special condition: missed flight (No-Show); Rate data: 50%; Calculation base object: ticket price.

[0036] These extracted entities are no longer part of the original text, but rather independent data units that have been tagged with standardized labels and can be directly processed by the program.

[0037] It should be noted that, in specific implementation scenarios, a context-aware dynamic prompt word guidance scheme can be adopted on the basis of the above solution. That is, when the artificial intelligence parsing engine performs entity extraction, it does not passively receive text, but can actively combine the ticket context information obtained in step S101 (such as the business type being "rebooking") and the preset prompt word library to dynamically generate a highly targeted "target prompt word". For example, the prompt word can explicitly instruct the engine: "Your current task is to process a rebooking request for an adult economy class ticket. Please focus on extracting all time windows, rates, and additional conditions related to the rebooking, and ignore descriptions about refunds." This context-aware guidance can significantly improve the accuracy and relevance of entity extraction.

[0038] A multi-granularity entity joint extraction scheme is adopted, meaning the engine can not only extract single entities but also simultaneously identify and establish logical relationships between entities. For example, it can not only identify the entities "50%" and "face value," but also clarify the "calculation relationship" between them—that is, "50% is the proportion applied to the face value." This relationship information is crucial for subsequently constructing the correct pricing logic. All of the above optional schemes fall within the protection scope of this application.

[0039] Step S103: The semantic entities are mapped into a matrix according to a preset multidimensional data model to generate a multidimensional rule matrix containing time windows and dynamic rates.

[0040] The purpose of this step is to transform discrete, unstructured semantic entities into a unified data structure with rigorous mathematical logic and computer addressability. In this step, the system no longer treats the extracted semantic entities (such as time descriptors, rates, etc.) as isolated information points, but rather organizes them organically according to a pre-designed multidimensional data model.

[0041] A multidimensional data model essentially defines the coordinate axes (dimensions) of a rule matrix. The two most critical dimensions are the time window and dynamic rates. Specific operations include time benchmark anchoring and standardization, binding rates to calculation logic, and integrating special conditional dimensions.

[0042] Time reference anchoring and standardization means using the "scheduled flight departure time" in the ticket context information as the absolute time calculation reference anchor point. All extracted time descriptors (such as "7 days before departure") will be converted into values ​​with uniform dimensions (e.g., all converted to minutes or hours) relative to this anchor point, and further constructed into a continuous time interval vector with explicit open and closed properties (such as [a,b)).

[0043] The rate is bound to the calculation logic, that is, each time interval vector is precisely bound to its corresponding rate data, rate representation type (absolute value or relative proportion), and calculation base object (such as ticket price).

[0044] Special condition dimension integration means that for special additional conditions such as no-show, they are integrated into the matrix as an independent Boolean dimension or state dimension, and the cost superposition logic when they are triggered is clearly defined (whether to replace or charge additional fees).

[0045] The resulting "multidimensional rule matrix" is a rigorously structured data container in which each element (or cell) precisely defines how to calculate refund and change fees under specific time intervals and conditions, thus providing a direct and unambiguous basis for subsequent automated pricing.

[0046] For example, suppose the flight is scheduled to depart at 14:00 on May 15, 2024. The system will perform matrix mapping on the extracted semantic entities: anchored to the time base, with the anchor point being 2024-05-15 14:00.

[0047] Construct time interval vectors: ">=7 days" is converted to a time interval vector: [-∞, -10080) minutes (i.e., from infinity to 10080 minutes before takeoff, 7 days = 10080 minutes); "[72 hours, 7 days)" is converted to: [-10080, -4320) minutes; "[24 hours, 72 hours)" is converted to: [-4320, -1440) minutes; "<24 hours" is converted to: [-1440, 0) minutes.

[0048] Binding rate and logic: Time interval [-10080, -4320), binding: rate 10%, type - relative proportion, base - ticket price, missed flight condition - no. Time interval [-1440, 0) binding: rate 50%, type - relative proportion, base - ticket price, missed flight condition - yes (and note that the superposition logic is "replace" the original interval fee).

[0049] Through the above mapping, all the scattered rule information is integrated into a multi-dimensional rule matrix with time as the horizontal axis and various business attributes (rate, base, special conditions, etc.) as the vertical axis (or depth dimension). When pricing is required, simply find the matching time interval based on the current time to directly read the complete billing rules.

[0050] It should be noted that, in specific implementation scenarios, a matrix nesting scheme for multiple flight segments / connecting tickets can be adopted based on the above solution. That is, for connecting tickets containing multiple billable flight segments, a sub-rule matrix can be generated for each independent billable flight segment, and these sub-matrices can be nested as higher-dimensional elements within a total multi-dimensional rule matrix. This can accurately handle complex scenarios where different flight segments have different refund and change rules. The above optional solutions fall within the scope of protection of this application.

[0051] Step S104: Perform logical confidence evaluation and cross-validation on the multidimensional rule matrix to obtain the validation results.

[0052] The core of this step lies in proactively identifying and intercepting potential risks caused by rule parsing errors or contradictions within the original rules themselves. After generating the multidimensional rule matrix, the system does not directly trust its content but instead initiates a rigorous, multidimensional, automated verification process.

[0053] Cross-validation mainly refers to checking the consistency of matrix content from different business logic perspectives. Typical validations include timeline integrity validation and business rationality validation.

[0054] The time axis integrity check maps all time windows in the matrix onto a one-dimensional linear time axis with the flight departure time as the origin. It checks for any time blind spots not covered by any rules, and for any logical conflicts where two or more rules overlap at the same time point and have inconsistent rates.

[0055] Business rationality verification involves judging the reasonableness of fare rate data by combining ticket context information (such as total ticket value). For example, if the fare rate is an absolute value, it is verified whether it exceeds the total ticket price; if the fare rate is a relative percentage, it is verified whether it is within a legal percentage range (such as 0%-100%).

[0056] Logical confidence assessment quantifies the credibility of the entire matrix based on cross-validation. The system assigns weights to each validation check and calculates a comprehensive "logical confidence score" based on the pass / fail status. The final "validation result" not only includes a simple "pass / fail" binary judgment but may also include specific error types and confidence scores, providing a basis for subsequent decision-making.

[0057] For example, suppose the multidimensional rule matrix generated by the system defines the following time windows (in minutes): [-∞,-10080): rate 0%; [-10080,-4320): rate 10%; [-4320,-1440): rate 20%; [-1440,0): rate 50%.

[0058] Cross-validation and timeline validation: The system found that the timeline from -∞ to 0 is completely covered with no blind spots; the boundaries of each interval are closely connected (e.g., -10080, -4320, etc.), with no overlap or conflict. This validation passed. Business rationality validation: The total face value of the tickets is 1000 yuan; the rates in the matrix are all relative proportions (0%, 10%, 20%, 50%), all within the legal range of 0%-100%. This validation passed. Logical confidence assessment: Since all preset validation items passed, the system assigns the matrix a high confidence score (e.g., 98 points) and generates "Validation result: Pass".

[0059] Counterexample: If the interval [-4320, -1440) is omitted from the matrix due to parsing errors, the time axis verification will find that [-4320, -1440) is a blind zone, and the verification result will be "failed", and the error type will be marked as "there is a time blind zone with undefined rates".

[0060] It should be noted that, in specific implementation scenarios, additional verification schemes for extreme business scenarios can be adopted based on the above solutions. That is, in addition to general verification, special verification rules can be designed for extreme scenarios unique to the aviation industry. For example, verifying whether there are any fee clauses under the "involuntary refund" rule; or verifying whether the tax refund policies of different countries / regions in the international ticket rules are logically consistent with the main ticket price rules.

[0061] A dynamic threshold adjustment scheme based on historical data is adopted, meaning the "confidence threshold" in the logical confidence assessment is not static. The system can dynamically learn and adjust the weights of various checks and the final confidence threshold based on historical manual review results. For example, if a specific type of low-confidence error has never historically led to an actual pricing error, the system can appropriately reduce the penalty weight for such errors. All of the above optional schemes fall within the scope of protection of this application.

[0062] Step S105: If the verification result is passed, then automatically calculate the price based on the multidimensional rule matrix, the ticket context information and the current system time, and output a structured refund and change fee result.

[0063] The core of this step lies in using rigorously validated structured rules to perform deterministic cost calculations within a precise time context. The system will only initiate this automated pricing process after the verification result of step S104 confirms the logical reliability of the multidimensional rule matrix, thereby ensuring the accuracy of the results and business security.

[0064] The specific execution process includes three key actions: time anchor matching, rule matrix query, and structured result generation. Time anchor matching: The system obtains the high-precision "current system time" (e.g., 2025-07-07T16:32:56+08:00) and combines it with the "scheduled flight departure time" in the ticket context information to calculate the precise time remaining until the flight's departure (e.g., 5 days and 3 hours remaining).

[0065] Rule matrix query: Using the calculated precise duration as the query key, it quickly locates the unique time window and all associated billing parameters (such as rate, base, special conditions, etc.) in the multi-dimensional rule matrix that has passed the verification.

[0066] Structured Result Generation: Based on the matched billing parameters and ticket context information (such as a total ticket value of 1000 yuan), the final pricing logic is executed, and the results are organized into a predefined, machine-readable, and front-end-friendly "structured data format". This format not only includes the final fee amount, but may also include metadata such as fee details, applicable rule summaries, fee types (handling fees / price differences, etc.), and currency units.

[0067] The entire process achieves an end-to-end automated closed loop from raw unstructured text to the final executable and displayable cost result.

[0068] It should be noted that, in specific implementation scenarios, a multi-factor composite pricing scheme can be adopted based on the above solution. That is, the automated pricing process can comprehensively consider factors from multiple dimensions. For example, in addition to the basic handling fee, if the rebooking involves a change in cabin class, the system can also automatically query the price difference between the old and new cabin classes, and combine the price difference with the handling fee to calculate a composite fee result that includes two sub-items: "handling fee" and "price difference".

[0069] A dynamic fee waiver determination scheme is adopted, which means that additional dynamic judgment logic can be introduced during the pricing process. For example, by combining the passenger membership level or specific promotional tags in the ticket context, even if the rule matrix indicates that a charge should be made, the system can dynamically adjust the final fee to zero according to a preset waiver strategy (such as waiver of handling fees for platinum card members), and indicate the reason for the waiver in the structured result. All of the above optional schemes are within the scope of protection of this application.

[0070] In some embodiments of this application, to improve the accuracy and efficiency of subsequent parsing, the method further includes the following steps before inputting the unstructured cancellation / rescheduling rule text into a preset artificial intelligence parsing engine: The unstructured refund / rescheduling rule text is preprocessed to remove irrelevant and redundant characters; Using a text segmentation algorithm, the preprocessed unstructured refund and rebooking rule text is divided into independent business logic segments to generate cleaned text blocks to be parsed. The business logic segments include at least a refund rule segment, a rebooking rule segment, and a missed flight supplementary rule segment.

[0071] As mentioned above, the original unstructured refund / change rule text is first preprocessed. This step aims to clean up noisy data in the text, specifically by removing irrelevant redundant characters, such as extra spaces, tabs, line breaks, special symbols (e.g., asterisks, underscores, and other typesetting marks), and HTML or XML tags. This operation yields a clean, coherent plain text, eliminating interference for subsequent semantic analysis.

[0072] Secondly, a text segmentation algorithm is used to logically divide the preprocessed text. Since complete refund and rebooking rules often mix descriptions of multiple business scenarios, inputting them as a whole into the parsing engine can easily lead to interference between different business logics, affecting the accuracy of entity extraction. Therefore, a segmentation strategy based on keywords, punctuation, or semantic coherence is adopted to divide the long text into multiple independent, semantically cohesive business logic paragraphs. These paragraphs should clearly distinguish at least three core categories: refund rule paragraphs (specifically describing refund conditions, time limits, and fees), rebooking rule paragraphs (specifically describing rebooking conditions, time limits, and fees), and missed flight supplementary rule paragraphs (specifically describing special handling rules and supplementary fees for passengers who fail to board their flights on time). The resulting cleaned text blocks, each focusing on a single business scenario, are fed into the AI ​​parsing engine as independent input units, ensuring clear context and a well-defined objective in the entity extraction process.

[0073] In some embodiments of this application, in order to provide high-quality structured semantic entities for subsequent steps, the unstructured refund / change rule text is input into a preset artificial intelligence parsing engine, specifically including: Identify the service type based on the ticket refund and change fee inquiry request; Based on the business type, target prompt words containing data extraction instructions and output specifications are dynamically assembled from a preset prompt word library; The target prompt, the cleaned text block, and the ticket context information are input into the artificial intelligence parsing engine.

[0074] As described above, firstly, the system identifies the specific business type involved based on the received ticket refund and change fee query request. This business type includes at least one of "refund," "change," or "missed flight processing," and the identification is based on the user's operational intent and the relevant status indicators in the ticket context information.

[0075] Secondly, based on the identified business type, the system dynamically assembles and generates a target prompt word from a pre-set prompt word library. This prompt word library pre-stores data extraction instruction templates and output format specifications for different business scenarios. The assembly process selects instruction content that matches the current business type (e.g., "Please extract all time limits and corresponding rates related to ticket refunds from the text") and structured output requirements (e.g., "Return in JSON format, with fields including time_window, fee_type, and fee_value"), thus forming a highly targeted and clearly defined target prompt word.

[0076] Finally, the target prompts, the pre-processed and cleaned text blocks (especially the business logic segments corresponding to the current business type, such as the refund rule segment used for refund requests), and the complete ticket context information (including flight number, cabin class, ticket price, departure time, etc.) are all fed into the AI ​​parsing engine as input. This multi-source information fusion method guides the parsing engine to accurately perform semantic understanding and key information extraction tasks within a specific business context, providing high-quality structured semantic entities for subsequent steps.

[0077] In some embodiments of this application, the semantic entities are matrix-mapped according to a preset multidimensional data model to generate a multidimensional rule matrix containing time windows and dynamic rates, specifically including: Extract time descriptors from the semantic entities and determine the scheduled flight departure time as a unified time calculation benchmark anchor point; The extracted time descriptors are converted into a unified time unit. Based on the logical relationship between the transformed time descriptor and the time calculation benchmark anchor point, a continuous time interval vector with opening and closing attributes is constructed.

[0078] As mentioned above, firstly, all time-related descriptors are extracted from the semantic entities output by the AI ​​parsing engine. These time descriptors are typically in natural language form, such as "7 days (inclusive) to 3 days (exclusive) before departure" or "within 48 hours of flight departure." Simultaneously, the system obtains the scheduled flight departure time corresponding to the ticket from the ticket context information and uses it as a unified time calculation benchmark for standardized calculations across all subsequent time intervals.

[0079] Secondly, the extracted time descriptors are converted into a unified unit of time measurement, typically using "minutes" or "seconds" as the standard unit. For example, "7 days" is converted to 10080 minutes, and "48 hours" is converted to 2880 minutes. This conversion process needs to take into account the ambiguous expressions in natural language (such as "inclusive," "exclusive," "before," and "after") and combine them with business conventions for precise quantification to ensure the consistency and comparability of time values.

[0080] Finally, based on the logical relationship between the converted time values ​​and the time calculation benchmark anchor point, a continuous time interval vector with explicit open and closed attributes is constructed. Specifically, taking the flight departure time as the time zero point (t=0), the time descriptor for "before departure" is mapped to a negative value interval. For example, "7 days (inclusive) to 3 days (exclusive) before departure" is mapped to a left-closed, right-open interval [−10080,−4320), where "inclusive of 7 days" corresponds to a closed left endpoint and "exclusive of 3 days" corresponds to an open right endpoint. In this way, time-rate pairs in multiple semantic entities are transformed one by one into structured time interval vectors with coherent beginning and end or verifiable coverage relationships, and bound to the corresponding dynamic rate parameters, ultimately forming a complete multidimensional rule matrix.

[0081] In some embodiments of this application, generating a multidimensional rule matrix that includes time windows and dynamic rates further includes: For each continuous time interval vector, extract the corresponding rate data, separate and record its numerical part and currency code; The rate data is determined to be represented by either an absolute value calculation or a relative proportion calculation. If the representation type is relative proportion calculation, then the calculation base object corresponding to the relative proportion calculation is further extracted. The calculation base object includes the ticket price, the actual ticket price, or the fare for the chargeable flight segment.

[0082] As described above, firstly, for each constructed continuous time interval vector, the associated fee data is extracted from the corresponding semantic entity. This fee data typically appears in natural language, such as "charge a 10% handling fee" or "refund fee is 200 yuan". The system needs to separate and record the numerical part (such as "10" or "20") and the currency code (such as "CNY" or "USD") separately. If the currency is not explicitly specified, the currency in the ticket context information is used by default.

[0083] Secondly, determine the representation type of the fee rate data, that is, whether it is an absolute value calculation or a relative proportion calculation. Absolute value calculation means that the fee is expressed as a fixed amount (such as "200 yuan"), while relative proportion calculation means that it is expressed as a percentage (such as "10%)". This determination is based on numerical suffixes, contextual keywords (such as "%", "proportion", "charged according to...", etc.), and common business rules.

[0084] If the determination result is a relative proportion calculation, it is necessary to further identify and extract the calculation base object corresponding to that proportion. The calculation base object is the reference benchmark for determining the final fee amount. According to aviation business practice, it mainly includes the following three categories: ticket price (i.e., the total ticket price shown on the ticket), actual ticket price (i.e., the amount actually paid by the passenger, which may include discounts or offers), or billable segment fare (i.e., the fare only for the specific segment where a refund or change operation occurs). This calculation base object is usually implicit in the rule text (e.g., "charged at 10% of the ticket price"), and needs to be accurately extracted through semantic analysis and bound together with the corresponding time interval and rate value to form a complete billing rule unit in the multi-dimensional rule matrix.

[0085] In some embodiments of this application, to achieve high-precision and automated calculation of cancellation and rebooking fees, a multi-dimensional rule matrix containing time windows and dynamic rates is generated, which further includes: Independently extract the additional conditions for missed flights and their corresponding time limits from the semantic entities; The system determines the additional fee stacking logic when the missed flight additional conditions are triggered. The additional fee stacking logic includes replacement logic that replaces the original time window fee, or stacking logic that adds a fee on top of the original time window fee. The continuous time interval vector, the rate data, the computational base object, and the cost superposition logic of the missed flight additional conditions are used as independent dimensions to assemble and generate the computer-addressable multidimensional rule matrix.

[0086] As mentioned above, firstly, the additional conditions related to "missing a flight" and their corresponding time limits are independently extracted from the semantic entities. Additional conditions for missing a flight typically manifest as rules such as "failure to process a refund within 2 hours of flight departure is considered a missed flight" or "an additional handling fee will be charged for refunds after missing a flight." The "time limit" is used to define whether a flight has been missed; for example, "2 hours before departure" is this limit. The system needs to parse it into a specific time point relative to the scheduled flight departure time and logically associate or distinguish it from the aforementioned continuous time interval vector.

[0087] Secondly, the system determines the corresponding fee surcharge logic when the additional conditions for missing a flight are triggered. This logic is divided into two categories: one is replacement logic, where the fee rules for missing a flight completely replace the original refund and rebooking fees for the original time window (for example, regardless of the original rules, a fixed high handling fee is charged once a flight is missed); the other is surcharge logic, where an additional fee is charged on top of the fees already determined for the original time window (for example, an additional 200 yuan fee is charged on top of the original 300 yuan refund fee). This determination is based on keywords in the rule text (such as "additional charge," "extra charge," "executed according to...", "no longer applicable to the original rules") and business semantics for accurate identification.

[0088] Finally, the aforementioned continuous time interval vector, fare data (including numerical values, currency codes, and representation types), calculation base objects (such as ticket face value, actual ticket price, or chargeable segment fare), and the fare stacking logic for missed flight conditions are all assembled as mutually orthogonal independent dimensions into a well-structured, complete, and directly addressable and invoked multidimensional rule matrix. This matrix uses time intervals as row indices and each fare element as a column dimension, enabling the subsequent fare calculation engine to quickly match and execute the corresponding fare calculation logic based on ticket status and operation time, achieving high-precision and automated calculation of refund and change fees.

[0089] In some embodiments of this application, to provide a high-confidence data foundation for subsequent automated calculation of refund and change fees, logical confidence evaluation and cross-validation are performed on the multidimensional rule matrix, specifically including: Map all time windows in the multidimensional rule matrix onto a one-dimensional linear time axis; Calculate the boundary values ​​of adjacent time windows and verify whether there are time blind spots with undefined rates in the time intervals on the one-dimensional linear time axis; Verify whether the same point in time is simultaneously included in two different time windows with different rates to detect logical overlap.

[0090] As described above, firstly, all continuous time interval vectors (i.e., time windows) in the multidimensional rule matrix are uniformly mapped onto a one-dimensional linear time axis with the scheduled flight departure time as the origin. This time axis is usually in minutes or seconds, with negative values ​​representing before departure and positive values ​​representing after departure, thereby achieving visual alignment and logical comparison of different rule entries under a unified time reference.

[0091] Secondly, for all adjacent time windows on the mapped timeline, their boundary values ​​(including the left and right endpoints) are extracted and sorted and analyzed for connectivity in chronological order. The connection between adjacent intervals is examined to determine if gaps exist. If there is an uncovered time period between the right endpoint of the preceding interval and the left endpoint of the following interval, this time period is identified as an "undefined rate time blind spot," indicating that the rule system lacks clear billing basis in this area, which may lead to business disputes or system anomalies.

[0092] Secondly, for each time point (especially the boundary points and key nodes within each time window), it is verified whether it falls within the range of two or more different time windows simultaneously. If the same time point is included in multiple windows, and the rate data (including values, representation types, or calculation bases) corresponding to these windows are inconsistent, it is determined as "logical overlap". Logical overlap will lead to uncertain cost calculation results, violate the rule uniqueness principle, and must be marked as a conflict item and trigger an alarm or manual review mechanism.

[0093] The above verification process constitutes the core logical verification of the integrity, consistency and executability of the multidimensional rule matrix, ensuring that the final generated rule system is fully covered, conflict-free and unambiguous in the time dimension, providing a high-confidence data foundation for subsequent automated calculation of refund and change fees.

[0094] In some embodiments of this application, to improve the logical confidence and practical usability of the entire refund and change billing rule system, the logical confidence evaluation and cross-validation of the multi-dimensional rule matrix are performed, further including: Obtain the total face value from the ticket context information; If the rate data in the multidimensional rule matrix is ​​represented as an absolute value, then verify whether the absolute value is less than or equal to the total face value. If the rate data in the multidimensional rule matrix is ​​represented as a relative proportion, then it is verified whether the relative proportion value is strictly distributed within the preset legal percentage closed interval.

[0095] As mentioned above, firstly, the total face value of the ticket is obtained from the ticket context information, that is, the total ticket price indicated on the ticket (usually excluding taxes and fees), which serves as a benchmark reference value for judging the reasonableness of the fee.

[0096] Secondly, for rate data represented by absolute values ​​in the multidimensional rule matrix, the system verifies whether the recorded amount is less than or equal to the total face value of the ticket. If a certain absolute rate (such as "refund fee of 5000 yuan") exceeds the total face value (such as "face value of 3000 yuan"), it is considered an unreasonable rule that may violate common business sense or consumer rights protection principles. It should be marked as an anomaly and a correction or review process should be triggered.

[0097] Secondly, for fee rate data represented as a relative proportion (such as "charge a 20% handling fee"), the system verifies whether its proportion value strictly falls within a preset legal percentage closed range. This legal range is usually set by business specifications or regulatory requirements, such as [0%, 100%] or more strictly [0%, 50%]. If the proportion value exceeds this range (such as "120%" or "-10%), it is determined to be an invalid rule, indicating a logical error or data parsing deviation, and should be blocked or alerted.

[0098] The aforementioned verification mechanism ensures that the rate parameters in the multi-dimensional rule matrix meet the requirements of business rationality, financial compliance, and system executability at the numerical level, thereby improving the logical confidence and actual usability of the entire refund and change billing rule system.

[0099] In some embodiments of this application, after obtaining the verification result, the method further includes: Based on the pass status of each cross-validation, a preset weighting algorithm is used to calculate the logical confidence score for the multidimensional rule matrix. If the logical confidence score is lower than the preset confidence threshold, or if a business extreme value conflict is captured in the cross-validation, the automated pricing process will be truncated. Generate an exception data snapshot containing an error type identifier, and downgrade the request to a manual review queue.

[0100] As described above, firstly, based on the pass status of each cross-validation (such as the presence of time blind spots, logical overlaps, rate exceeding limits, and proportion exceeding limits), a comprehensive logical confidence score is calculated on the multi-dimensional rule matrix using a pre-defined weighting algorithm. This weighting algorithm assigns different weights according to the degree of impact of each validation item on business risk; for example, "time blind spots" and "logical overlaps" may be given higher weights, while "slight proportion exceeding limits" has a lower weight. The weighted summation of the scores for all validation items forms the final logical confidence score, used to quantify the credibility of the rule matrix.

[0101] Secondly, if the confidence score of the logic is lower than the preset confidence threshold (e.g., set to 90 points), or if a business extreme value conflict is captured during the cross-validation process (e.g., the refund fee is higher than the total ticket value, the cost after missing the flight is negative, or there are irreconcilable rate conflicts at the same time point, etc., serious logical errors), the system will immediately cut off the current automated pricing process to prevent the generation of erroneous or unexplainable cost results.

[0102] Finally, the system automatically generates an anomaly data snapshot containing an error type identifier. This snapshot fully records the rule entry that triggered the anomaly, its corresponding dimension (such as time window, rate type, missed flight conditions, etc.), the specific error category (such as "time blind spot," "rate overrun," "logical overlap," etc.), and the original ticket context information. Subsequently, the current refund / change request, along with this snapshot, is downgraded and routed to a manual review queue, where professionals intervene to verify and process it, ensuring that business compliance and user experience are not affected.

[0103] In some embodiments of this application, to improve cost transparency and user trust, automated pricing is performed based on the multidimensional rule matrix, the ticket context information, and the current system time, outputting structured refund and change fee results, specifically including: Calculate the precise time interval vector that the current system time matches in the multidimensional rule matrix; The rate rules corresponding to the precise time interval vector are matched with the calculation base object to calculate the price, and when there is a missed flight indicator, the additional fee is calculated together to generate a precise refund and rebooking fee value; The precise refund and change fee values ​​and details are converted into standard data interaction messages and sent to the downstream ticketing settlement system, and explanatory text is generated and fed back to the client.

[0104] As described above, firstly, the current system time (usually represented as a timestamp accurate to the second in UTC or the local time zone) is obtained and compared with all continuous time interval vectors in the multidimensional rule matrix to determine the precise time interval vector that the current time matches. This matching process must strictly adhere to the inclusion relationship on the time axis to ensure that only one valid interval is matched; if multiple matches or no matches occur due to insufficient verification, it is considered an anomaly and the aforementioned degradation mechanism is triggered.

[0105] Secondly, after hitting the precise time interval vector, the system extracts the associated rate rules (including rate values ​​and whether the representation type is absolute or relative) and the corresponding calculation base object (such as ticket face value, actual ticket price, or fare for a specific flight segment) for that interval. Based on the representation type, the corresponding pricing logic is executed: if it is an absolute value, the amount is directly used; if it is a relative proportion, it is multiplied by the calculation base object to obtain the cost value. Simultaneously, the system checks whether there is a missed flight indicator in the ticket context information (such as failure to process refunds or changes within the specified time limit). If it exists, the additional costs are combined and included according to the preset missed flight supplementary conditions and their cost aggregation logic (replacement or aggregation) in the multi-dimensional rule matrix, ultimately generating the precise refund and change cost value.

[0106] Finally, the system encapsulates the precise refund and change fee amount and its detailed breakdown (including base fee, no-show surcharge, calculation basis, applicable time range, fee type, etc.) into a structured data exchange message (such as JSON or XML format) conforming to industry standards or internal company specifications, and sends it to the downstream ticketing settlement system for subsequent accounting processing and voucher generation. Simultaneously, the system automatically generates explanatory text feedback for the user (e.g., "You applied for a refund within one hour of your flight departure, which is subject to the no-show rule; the refund fee is 50% of the ticket price plus a surcharge of 200 yuan, totaling 800 yuan"), and returns this information through the client interface or API, enhancing fee transparency and user trust.

[0107] Compared with existing technologies, this application discloses an intelligent structured parsing method for airline refund and rebooking rules. This method achieves high-precision and robust automated rule parsing and pricing. By inputting unstructured refund and rebooking rule text into a preset artificial intelligence parsing engine to extract semantic entities, and further mapping these entities into a matrix according to a preset multidimensional data model, a multidimensional rule matrix containing time windows and dynamic rates is generated. This successfully transforms fuzzy natural language rules into a low-level data structure with strict mathematical logic that can be directly addressed and calculated by computers. This process fundamentally solves the problems of accuracy and reliability in rule parsing.

[0108] By eliminating the financial and compliance risks associated with model "illusions," the solution introduces key steps such as confidence assessment and cross-validation of the execution logic of the multi-dimensional rule matrix. This step proactively detects and intercepts unreasonable results (such as time blind spots, logical overlaps, and fees exceeding ticket prices) caused by model parsing errors or contradictions in the original rules. Subsequent automated pricing is only permitted if the validation result passes. This closed-loop architecture of "parsing-validation-execution" effectively isolates the risks between non-deterministic AI output and deterministic financial settlement, ensuring the safety and reliability of the final output of refund and change fees.

[0109] This solution constructs an efficient business processing closed loop and intelligent degradation mechanism. It not only achieves millisecond-level automated pricing, greatly improving processing efficiency, but also establishes a robust fault-tolerance system through a confidence assessment mechanism. When verification fails or confidence is insufficient, the system can automatically generate an abnormal snapshot and degrade the request to a manual review queue. This avoids erroneous pricing, ensures a good user experience, and forms a highly robust and complete business closed loop.

[0110] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0111] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0112] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0113] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A smart structured parsing method for airline refund and change rules, characterized in that, include: Receive a ticket refund and change fee query request sent by the client, and obtain the unstructured refund and change rule text and ticket context information corresponding to the request; The unstructured refund and rescheduling rule text is input into a preset artificial intelligence parsing engine to extract semantic entities related to refunds and rescheduling; The semantic entities are mapped into a matrix according to a preset multidimensional data model to generate a multidimensional rule matrix containing time windows and dynamic rates. Perform logical confidence evaluation and cross-validation on the multidimensional rule matrix to obtain the validation results; If the verification result is successful, the price is automatically calculated based on the multidimensional rule matrix, the ticket context information, and the current system time, and a structured refund and change fee result is output.

2. The method as described in claim 1, characterized in that, Before inputting the unstructured refund / refund rule text into a preset artificial intelligence parsing engine, the method further includes: The unstructured refund / rescheduling rule text is preprocessed to remove irrelevant and redundant characters; Using a text segmentation algorithm, the preprocessed unstructured refund and rebooking rule text is divided into independent business logic segments to generate cleaned text blocks to be parsed. The business logic segments include at least a refund rule segment, a rebooking rule segment, and a missed flight supplementary rule segment.

3. The method as described in claim 2, characterized in that, The unstructured refund / rescheduling rule text is input into a preset artificial intelligence parsing engine, specifically including: Identify the service type based on the ticket refund and change fee inquiry request; Based on the business type, target prompt words containing data extraction instructions and output specifications are dynamically assembled from a preset prompt word library; The target prompt, the cleaned text block, and the ticket context information are input into the artificial intelligence parsing engine.

4. The method as described in claim 1, characterized in that, The step of mapping the semantic entities according to a preset multidimensional data model to generate a multidimensional rule matrix containing time windows and dynamic rates specifically includes: Extract time descriptors from the semantic entities and determine the scheduled flight departure time as a unified time calculation benchmark anchor point; The extracted time descriptors are converted into a unified time unit. Based on the logical relationship between the transformed time descriptor and the time calculation benchmark anchor point, a continuous time interval vector with opening and closing attributes is constructed.

5. The method as described in claim 4, characterized in that, The generation of the multidimensional rule matrix, which includes time windows and dynamic rates, also includes: For each continuous time interval vector, extract the corresponding rate data, separate and record its numerical part and currency code; The rate data is determined to be represented by either an absolute value calculation or a relative proportion calculation. If the representation type is relative proportion calculation, then the calculation base object corresponding to the relative proportion calculation is further extracted. The calculation base object includes the ticket price, the actual ticket price, or the fare for the chargeable flight segment.

6. The method as described in claim 5, characterized in that, The generation of the multidimensional rule matrix, which includes time windows and dynamic rates, also includes: Independently extract the additional conditions for missed flights and their corresponding time limits from the semantic entities; The system determines the additional fee stacking logic when the missed flight additional conditions are triggered. The additional fee stacking logic includes replacement logic that replaces the original time window fee, or stacking logic that adds a fee on top of the original time window fee. The continuous time interval vector, the rate data, the computational base object, and the cost superposition logic of the missed flight additional conditions are used as independent dimensions to assemble and generate the computer-addressable multidimensional rule matrix.

7. The method as described in claim 1, characterized in that, The step of performing logical confidence evaluation and cross-validation on the multidimensional rule matrix specifically includes: Map all time windows in the multidimensional rule matrix onto a one-dimensional linear time axis; Calculate the boundary values ​​of adjacent time windows and verify whether there are time blind spots with undefined rates in the time intervals on the one-dimensional linear time axis; Verify whether the same point in time is simultaneously included in two different time windows with different rates to detect logical overlap.

8. The method as described in claim 7, characterized in that, The step of performing logical confidence evaluation and cross-validation on the multidimensional rule matrix further includes: Obtain the total face value from the ticket context information; If the rate data in the multidimensional rule matrix is ​​represented as an absolute value, then verify whether the absolute value is less than or equal to the total face value. If the rate data in the multidimensional rule matrix is ​​represented as a relative proportion, then it is verified whether the relative proportion value is strictly distributed within the preset legal percentage closed interval.

9. The method as described in claim 8, characterized in that, After obtaining the verification result, the process also includes: Based on the pass status of each cross-validation, a preset weighting algorithm is used to calculate the logical confidence score for the multidimensional rule matrix. If the logical confidence score is lower than the preset confidence threshold, or if a business extreme value conflict is captured in the cross-validation, the automated pricing process will be truncated. Generate an exception data snapshot containing an error type identifier, and downgrade the request to a manual review queue.

10. The method as described in claim 1, characterized in that, The automated pricing based on the multidimensional rule matrix, the ticket context information, and the current system time, outputting structured refund and change fee results, specifically includes: Calculate the precise time interval vector that the current system time matches in the multidimensional rule matrix; The rate rules corresponding to the precise time interval vector are matched with the calculation base object to calculate the price, and when there is a missed flight indicator, the additional fee is calculated together to generate a precise refund and rebooking fee value; The precise refund and change fee values ​​and details are converted into standard data interaction messages and sent to the downstream ticketing settlement system, and explanatory text is generated and fed back to the client.

Images