Automatic civil aviation air ticket booking system based on semantic recognition

Abstract

The invention discloses a civil aviation air ticket automatic booking system based on semantic recognition, and the system comprises a booking input data obtaining module which obtains booking input data; a semantic understanding engine that performs semantic recognition processing; the large language model analysis module is used for performing semantic representation, context reasoning and semantic confidence generation; pre-training a language model semantic coding module, and executing semantic coding; the user portrait and strategy analysis module is used for generating a user portrait feature vector; the multi-dimensional intelligent decision-making module is used for generating a comprehensive recommendation score; the unified data interface adaptation module is used for sending a seat reservation request to a corresponding system and processing an abnormal condition; the automatic ticket issuing module is used for completing automatic ticket issuing and executing a predetermined failure rollback process when ticket issuing fails; and the predetermined log management module is used for performing log recording on the result. According to the method, artificial intelligence technologies such as semantic recognition, a large language model, user portrait analysis and automatic decision making are combined, and full automation of the civil aviation air ticket booking process is achieved.

Description

Technical Field

[0001] This invention relates to the fields of civil aviation ticketing processing and artificial intelligence semantic recognition technology, and in particular to an automatic civil aviation ticket booking system based on semantic recognition. Background Technology

[0002] With the continuous improvement of informatization in the civil aviation industry, online ticketing systems have become the main way for passengers to obtain flight information and complete ticket purchases. However, existing civil aviation ticketing systems still rely on manual input of structured fields, such as departure point, destination, date, number of passengers, airline preferences, and cabin class selection. Users often need to click, filter, and verify multiple times to complete a full booking and ticketing process. For complex scenarios, such as multiple travelers, cross-airline price comparisons, historical preference matching, and fuzzy time period expressions, existing systems generally lack intelligent processing capabilities. At the same time, passengers often describe their needs in natural language when actually booking tickets, such as "Book me a flight to Beijing tomorrow afternoon" or "Choose a cheap flight that isn't too late." Traditional structured retrieval methods struggle to accurately understand such semantic expressions, resulting in a poor user experience.

[0003] While existing technologies include natural language processing-based ticketing assistance tools, most remain at the keyword matching level and cannot utilize deep semantic understanding models to perform complex intent parsing. Furthermore, they cannot integrate semantic parsing results with real-time flight data, seat inventory data, and user historical preferences for joint decision-making. In addition, current reservation systems generally rely on manual or semi-automatic methods to construct reservation requests, lacking a unified data format and instruction conversion mechanism. Different airlines and Global Distribution Systems (GDS) use different protocols, such as NDCXML, Eterm commands, and Amadeus / Sabre interfaces. Existing systems typically require manual adaptation, which is not only inefficient but also prone to problems such as inventory inconsistencies, parsing errors, and communication failures. Existing technologies also struggle to achieve automatic ticketing after successful reservations. The ticketing process involves multiple steps, including PNR acquisition, ticket number generation, status backfilling, and rollback in case of ticketing failure; traditional systems have not formed a truly automated closed loop.

[0004] On the other hand, while some platforms attempt to use users' historical order data for recommendations, their analytical dimensions are limited. They can typically only sort users by price or time, failing to combine multi-dimensional features such as price sensitivity, airline preference, cabin class preference, and time period preference to generate user profiles. Furthermore, they cannot implement intelligent, multi-weighted comprehensive scoring models in the booking decision-making process. This makes it difficult for the system to provide personalized booking solutions based on users' actual needs, effectively reduce labor costs, and meet the growing demand for automated ticketing.

[0005] Existing technologies generally suffer from insufficient semantic parsing capabilities, incomplete user profiling, lack of unified interface adaptation capabilities, absence of automated decision-making models, and inability to automate the ticketing process, making it difficult to form a fully automated chain from semantic input to seat reservation and ticketing. Therefore, how to provide a civil aviation ticket automatic reservation system based on semantic recognition is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an automated civil aviation ticket booking system based on semantic recognition. This invention fully utilizes technologies such as large language model parsing, semantic encoding, multi-dimensional intelligent recommendation, unified data adaptation, and automated ticketing, and details a fully automated closed-loop solution from parsing the user's natural language booking request to final ticketing completion. The system uses a semantic understanding engine and pre-trained language models to recognize user intent and extract structured elements. Combined with user profiling and historical preference analysis modules, it models user booking behavior. A multi-dimensional intelligent decision-making module then comprehensively scores and ranks multiple candidate flights, automatically generating the optimal booking plan. Through a unified data interface adaptation module, it seamlessly connects with the target airline or global distribution system to complete booking and ticketing. This system boasts advantages such as strong understanding capabilities, fast response speed, high recommendation accuracy, high degree of automation in the operation process, and comprehensive anomaly handling capabilities, significantly improving the intelligence level and user experience of online ticketing.

[0007] An automatic civil aviation ticket booking system based on semantic recognition according to an embodiment of the present invention includes:

[0008] The reservation input data acquisition module is used to acquire the natural language ticket booking request and related passenger data entered by the user through the terminal, and form reservation input data;

[0009] A semantic understanding engine is used to perform semantic recognition processing on natural language ticket booking requests and generate structured booking feature vectors.

[0010] The large language model parsing module is used to perform semantic representation, contextual reasoning, and semantic confidence generation on natural language ticket booking requests.

[0011] The pre-trained language model semantic encoding module is used to perform semantic encoding on natural language ticket booking requests and generate intermediate semantic vectors;

[0012] The user profiling and strategy analysis module is used to perform statistics and feature extraction on historical flight selection records, cabin class selection records, price range records, airline selection records, and time period preference data based on users' historical order data, and generate user profile feature vectors.

[0013] The multidimensional intelligent decision-making module is used to input the structured booking feature vector and the user profile feature vector into the comprehensive scoring model. Based on the time matching degree score, user preference matching score, historical behavior similarity score and cabin availability score, it generates a comprehensive recommendation score and performs joint decision-making with semantic confidence to determine standardized flight booking requests.

[0014] The unified data interface adaptation module is used to perform multi-source data parsing, field mapping, and protocol conversion on standardized flight booking requests, send booking requests to the corresponding system, and handle abnormal situations.

[0015] The automatic ticketing module is used to construct a ticketing request message after a successful reservation, send a ticketing request according to the ticketing interface protocol, parse the ticketing response, and write the reservation record code and electronic ticket number into the order record. When ticketing fails, it executes a reservation failure rollback process.

[0016] The reservation log management module is used to log reservation input data, semantic recognition results, user profile feature vectors, decision results, request and response messages during the interface adaptation phase, and reservation and ticketing results.

[0017] Optionally, modules can be integrated using the following methods:

[0018] Step 1: Obtain the natural language booking request entered by the user through the terminal and related passenger data to form the booking input data;

[0019] Step 2: Perform semantic recognition processing on the natural language booking requests in the pre-booking input data, match the booking elements with real-time flight information and seat inventory data, and perform field mapping and structuring processing to generate a structured booking feature vector;

[0020] Step 3: Perform statistical analysis and feature extraction on historical order data to generate a user preference feature set, and combine the user preference feature set in a unified format to form a user profile feature vector;

[0021] Step 4: Input the structured booking feature vector and user profile feature vector into the decision calculation process, calculate the comprehensive recommendation score, and make a joint decision with the semantic confidence obtained from semantic recognition to determine the standardized flight booking request;

[0022] Step 5: Perform protocol adaptation processing on standardized flight booking requests, send booking requests to the target system, and handle exceptions such as communication timeouts, inventory inconsistencies, and message parsing errors.

[0023] Step Six: After receiving the response data indicating a successful reservation, execute the ticketing process; if ticketing fails or an inventory discrepancy is detected, execute the reservation failure rollback process.

[0024] Optionally, step one specifically includes:

[0025] Based on the natural language booking request, a flight query request is sent to the preset flight data service to obtain real-time flight information that matches the natural language booking request from the flight data service;

[0026] In accordance with the internal reservation data format specifications, the fields of natural language ticketing requests, relevant passenger data, real-time flight information and seat inventory data were aligned and aggregated.

[0027] Generate a corresponding reservation identifier and time identifier for each reservation service, and organize the aggregated data into a unified structure to form reservation input data;

[0028] Establish a correspondence between log identifiers and predefined identifiers, and write the predefined input data into the log record.

[0029] Optionally, step two specifically includes:

[0030] Read the text field corresponding to the natural language ticket booking request from the pre-defined input data, perform character normalization and segmentation on the text field, and obtain the natural language text to be parsed;

[0031] Perform semantic encoding on natural language text to generate intermediate semantic vectors;

[0032] Context parsing is performed based on intermediate semantic vectors to determine the semantic relationships and syntactic structure in natural language ticket booking requests, and the context parsing results are obtained.

[0033] Based on the context parsing results, intent extraction is performed to identify booking elements related to ticket booking;

[0034] According to the internal structured coding rules, the corresponding reservation elements, real-time flight information fields and seat inventory data fields are combined and coded to generate a structured reservation feature vector.

[0035] The intermediate semantic vectors, context parsing results, pre-defined elements, and structured pre-defined feature vectors corresponding to the current pre-defined business generated during semantic recognition processing are written into the log records, and associated with the corresponding pre-defined identifiers and log identifiers.

[0036] Optionally, step three specifically includes:

[0037] Retrieves user's historical order data from the historical order database based on user identifier;

[0038] Based on the sorted historical order data, the frequency of user selection on different flights is counted to generate a user's historical flight selection record;

[0039] Based on the sorted historical order data, the number of times and percentage of users selected each cabin class are statistically analyzed to generate a record of users' historical cabin class selections.

[0040] Based on the sorted historical order data, the number of orders and price distribution ranges of users in different price ranges are statistically analyzed to generate a record of users' historical price ranges.

[0041] Based on the sorted historical order data, the number and proportion of orders placed by users on different airlines are statistically analyzed to generate a record of users' historical airline selections. The distribution of users' departure times in different time periods is statistically analyzed, and the departure times are mapped to preset time periods to generate users' historical time period preference data.

[0042] Based on the user's historical price range records, the proportion of the user's selection in each price range is calculated to generate the user's price sensitivity characteristics;

[0043] Calculate the selection ratio of each time period interval based on users' historical time period preference data to generate time period preference features;

[0044] Based on the user's historical cabin class selection records, the selection ratio of each cabin class is calculated to generate cabin class preference features;

[0045] Calculate the selection ratio of each airline based on the user's historical airline selection records to generate airline preference features;

[0046] User price sensitivity features, time period preference features, cabin class preference features and airline preference features are organized according to the preset field order and encoding method to form a user preference feature set, and the user preference feature set is normalized.

[0047] The processed user preference feature set is mapped to a numerical vector of a set dimension to generate a user profile feature vector;

[0048] Write the user's historical order data summary, user preference feature set, and user profile feature vector generated from the user preference feature set into the log record, and record the correspondence between user identifier and reservation identifier.

[0049] Optionally, step four specifically includes:

[0050] A time matching score is calculated based on the matching relationship between the departure and arrival times of candidate flights and the user's time preferences.

[0051] A preference matching score is calculated based on the matching relationship between the airline and cabin class of the candidate flights and the airline preference features and cabin class preference features in the user preference feature set.

[0052] A historical behavior similarity score is calculated based on the similarity between the candidate flight and the user's historical flight selection records.

[0053] A cabin space availability score is calculated based on the relationship between the number of remaining seats and a preset inventory threshold.

[0054] Based on preset weights, the price score, time matching score, user preference matching score, historical behavior similarity score, and cabin availability score are weighted and calculated to obtain the comprehensive recommendation score for each candidate flight.

[0055] The candidate flights are ranked, and the candidate flights and their corresponding cabin classes that meet the preset conditions in terms of both comprehensive recommendation score and semantic confidence are selected as the target flights and target cabin classes.

[0056] Based on the target flight and target cabin class, extract the field information of the target flight from the structured booking feature vector;

[0057] According to the internal booking instruction coding standard, the field information is organized into a unified format to generate a standardized flight booking request;

[0058] The price score, time matching score, user preference matching score, historical behavior similarity score, cabin availability score, comprehensive recommendation score, semantic confidence score, target flight and target cabin information, and the generated standardized ticket booking request for each candidate flight are written into the log record.

[0059] Optionally, step five specifically includes:

[0060] Based on standardized flight booking requests, the target booking system is determined to be either the airline's system or the global distribution system;

[0061] And select the corresponding protocol adaptation rule based on the target system identifier;

[0062] According to the protocol adaptation rules, the internally unified data format is parsed to obtain an intermediate data structure that is consistent with the reservation interface fields of the target system.

[0063] Based on the intermediate data structure and the target system's protocol specifications, a reservation instruction format is generated:

[0064] The generated reservation instruction format is encapsulated into a communication request message and sent to the corresponding airline system or global distribution system through a preset communication channel;

[0065] During the process of sending reservation requests and receiving reservation responses, the communication status is monitored. If a reservation response is not received after a preset time, the current situation is determined to be a communication timeout.

[0066] The reservation request is resent according to the preset retry rules. If the reservation request is still unsuccessful after the number of retries is exceeded, the reservation request is marked as a communication failure.

[0067] During the parsing of reservation response data, the message structure, number of fields, and field types are detected. When a message format that does not conform to the target system protocol specification is detected, the current situation is judged as a message parsing error, the corresponding error type and error field information are recorded, and the associated reservation request is marked as parsing failure.

[0068] After parsing, the inventory information returned in the reservation response data is compared with the corresponding seat inventory data in the structured reservation feature vector;

[0069] When an inconsistency is detected in the inventory quantity or the availability of goods, the current situation is judged as an inventory inconsistency, an inventory inconsistency flag is generated, and the corresponding reservation request is included in the reservation failure handling process.

[0070] The reservation instruction format, reservation request communication message, target system identifier, communication status, reservation response data, and exception judgment results such as communication timeout, inventory inconsistency, and message parsing error generated during the protocol adaptation phase are written into the log record.

[0071] Optionally, step six specifically includes:

[0072] Based on the reservation result status field in the reservation response data, the reservation result is judged. When the reservation result status is successful, the reservation record code is extracted from the reservation response data.

[0073] Based on the associated standardized flight booking request and reservation record code, generate ticketing request data according to the ticketing interface specification, and encapsulate the generated ticketing request data into a ticketing request message.

[0074] The ticketing request message is sent to the corresponding airline system or global distribution system through a pre-set communication channel;

[0075] The ticketing response message is parsed according to the ticketing interface protocol to extract the ticketing result status, reservation record code, and electronic ticket number;

[0076] When the ticketing result status is successful, the parsed reservation record code and electronic ticket number are written into the corresponding order record, the order status field in the order record is updated to the ticketed status, and the ticketing completion time and ticketing channel information are recorded.

[0077] When the ticketing result status is failed, or an inventory inconsistency flag, message structure error, or field missing is detected, the current ticketing request is marked as ticketing failure, and the reason for failure is written into the failure reason field in the order record.

[0078] The current ticketing request is marked as a ticketing failure time, and the booking failure rollback process is initiated, generating a rollback instruction according to the preset rules of the booking failure rollback process;

[0079] When the rollback rule requires cancellation of a reservation, a cancellation request message containing the reservation record code is constructed and sent to the corresponding airline system or global distribution system.

[0080] When canceling a reservation fails, mark the current cancellation operation as a rollback failure and record the reason for the rollback failure;

[0081] The ticketing request message, ticketing response message, reservation record code, electronic ticket number, order status update result, and the triggering conditions, rollback instruction sending record, and rollback result of the reservation failure rollback process are written to the log record.

[0082] The beneficial effects of this invention are:

[0083] This invention constructs an automatic civil aviation ticket booking system based on semantic recognition, achieving intelligent upgrades in several key aspects such as natural language understanding, personalized recommendation, data interface adaptation, and automatic ticketing, significantly improving booking efficiency and accuracy.

[0084] This invention introduces a semantic understanding engine and a pre-trained language model semantic encoding module, which can accurately parse users' booking requests input in natural language, extract key elements such as departure point, destination, date, and cabin class preference, and generate structured booking feature vectors. This effectively solves the problem of insufficient parsing ability of traditional systems when dealing with fuzzy expressions and non-standard sentences.

[0085] Through the user profiling and strategy analysis module, the system aggregates and models users' historical behavior, extracting multi-dimensional preference features such as price sensitivity, time period preference, cabin class preference, and airline selection, and generating user profile feature vectors. Based on this, the multi-dimensional intelligent decision-making module can combine real-time flight information and inventory status for accurate scoring and intelligent ranking, thereby generating the booking plan that best matches user preferences, significantly improving recommendation accuracy and user satisfaction.

[0086] The unified data interface adaptation module can automatically adapt to the interface specifications of various airline systems or GDS platforms based on standardized booking requests, realizing format conversion and data mapping. It also has the ability to handle abnormal situations such as communication timeouts, inventory inconsistencies, and message structure errors, ensuring the stability and robustness of the system. The automatic ticketing module further realizes a closed-loop process from reservation to ticketing, supports automatic rollback in case of ticketing failure, and ensures the integrity and consistency of ticketing processing.

[0087] This invention realizes intelligent processing of the entire process from natural language ticket booking to automatic ticketing, which not only improves the automation level and fault tolerance of the system, but also significantly optimizes the user experience and operational efficiency, and has high practical value and promotion prospects. Attached Figure Description

[0088] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0089] Figure 1 This is a schematic diagram of the structure of an automatic civil aviation ticket reservation system based on semantic recognition proposed in this invention;

[0090] Figure 2 This is an overall flowchart of a method for automatically booking civil aviation tickets based on semantic recognition proposed in this invention. Detailed Implementation

[0091] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0092] refer to Figures 1-2 An automated civil aviation ticket booking system based on semantic recognition includes:

[0093] The reservation input data acquisition module is used to acquire the natural language ticket booking request and related passenger data entered by the user through the terminal, and form reservation input data;

[0094] A semantic understanding engine is used to perform semantic recognition processing on natural language ticket booking requests and generate structured booking feature vectors.

[0095] The large language model parsing module is used to perform semantic representation, contextual reasoning, and semantic confidence generation on natural language ticket booking requests.

[0096] The pre-trained language model semantic encoding module is used to perform semantic encoding on natural language ticket booking requests and generate intermediate semantic vectors;

[0097] The user profiling and strategy analysis module is used to perform statistics and feature extraction on historical flight selection records, cabin class selection records, price range records, airline selection records, and time period preference data based on users' historical order data, and generate user profile feature vectors.

[0098] The multidimensional intelligent decision-making module is used to input the structured booking feature vector and the user profile feature vector into the comprehensive scoring model. Based on the time matching degree score, user preference matching score, historical behavior similarity score and cabin availability score, it generates a comprehensive recommendation score and performs joint decision-making with semantic confidence to determine standardized flight booking requests.

[0099] The unified data interface adaptation module is used to perform multi-source data parsing, field mapping, and protocol conversion on standardized flight booking requests, send booking requests to the corresponding system, and handle abnormal situations.

[0100] The automatic ticketing module is used to construct a ticketing request message after a successful reservation, send a ticketing request according to the ticketing interface protocol, parse the ticketing response, and write the reservation record code and electronic ticket number into the order record. When ticketing fails, it executes a reservation failure rollback process.

[0101] The reservation log management module is used to log reservation input data, semantic recognition results, user profile feature vectors, decision results, request and response messages during the interface adaptation phase, and reservation and ticketing results.

[0102] In this embodiment, the modules are connected through the following method:

[0103] Step 1: Obtain the natural language booking request entered by the user through the terminal and related passenger data to form the booking input data;

[0104] Step 2: Perform semantic recognition processing on the natural language booking requests in the pre-booking input data, match the booking elements with real-time flight information and seat inventory data, and perform field mapping and structuring processing to generate a structured booking feature vector;

[0105] Step 3: Perform statistical analysis and feature extraction on historical order data to generate a user preference feature set, and combine the user preference feature set in a unified format to form a user profile feature vector;

[0106] Step 4: Input the structured booking feature vector and user profile feature vector into the decision calculation process, calculate the comprehensive recommendation score, and make a joint decision with the semantic confidence obtained from semantic recognition to determine the standardized flight booking request;

[0107] Step 5: Perform protocol adaptation processing on standardized flight booking requests, send booking requests to the target system, and handle exceptions such as communication timeouts, inventory inconsistencies, and message parsing errors.

[0108] Step 6: After receiving the response data indicating a successful reservation, execute the ticketing process; if ticketing fails or an inventory discrepancy is detected, execute the reservation failure rollback process.

[0109] In this embodiment, step one specifically includes:

[0110] The natural language booking request includes departure city, arrival city, travel date, time preference, number of people, and cabin class preference;

[0111] The relevant passenger data includes at least one of the following: passenger name, document type, document number, and contact information. The format and integrity of the natural language booking request and the relevant passenger data are verified.

[0112] Based on the departure city, arrival city, travel date and time preference information contained in the natural language booking request, a flight query request is sent to the preset flight data service to obtain real-time flight information that matches the natural language booking request from the flight data service.

[0113] The real-time flight information includes flight number, departure airport, arrival airport, scheduled departure time, and scheduled arrival time.

[0114] Based on the flight number in the real-time flight information, a seat inventory query request is sent to the seat inventory service to obtain the seat inventory data corresponding to each candidate flight.

[0115] The seat inventory data includes cabin class, number of remaining seats, and availability for sale.

[0116] In accordance with the internal reservation data format specifications, the natural language ticketing requests, relevant passenger data, real-time flight information and seat inventory data are aligned and aggregated to generate corresponding reservation identifiers and time identifiers for each reservation business. The aggregated data is then organized into a unified structure to form reservation input data.

[0117] Establish a correspondence between log identifiers and reservation identifiers, write reservation input data into the log record, and the reservation input data includes at least the original text of the natural language ticket booking request, relevant passenger data, reservation identifiers and time identifiers, and record the data aggregation completion time.

[0118] In this embodiment, step two specifically includes:

[0119] Read the text field corresponding to the natural language ticket booking request from the pre-defined input data, perform character normalization and segmentation on the text field, and obtain the natural language text to be parsed;

[0120] Perform semantic encoding on natural language text to generate an intermediate semantic vector that represents the semantic information of a natural language ticket booking request;

[0121] The semantic encoding is implemented through a pre-trained language model semantic encoding algorithm, a semantic representation generation algorithm based on context semantic modeling, or a semantic feature extraction algorithm based on attention weight allocation, and is used to convert natural language text into an intermediate semantic vector representing semantic information.

[0122] Context parsing is performed based on intermediate semantic vectors to determine the semantic relationships and syntactic structure in natural language ticket booking requests, and the context parsing results are obtained.

[0123] The context parsing is achieved through parsing algorithms based on contextual semantic relationship modeling, semantic relationship parsing algorithms based on dependency parsing, or syntactic structure recognition algorithms based on semantic role labeling. It is used to determine the semantic relationships and syntactic structures in natural language ticket booking requests and obtain context parsing results.

[0124] Based on the context parsing results, intent extraction is performed to identify booking elements related to ticket booking, including departure city, arrival city, travel date, time preference, number of passengers, and cabin class requirements.

[0125] The reservation elements are matched and aligned with the departure airport, arrival airport, flight date, and departure time fields in the real-time flight information, as well as the cabin class and number of seats fields in the seat inventory data.

[0126] According to the internal structured coding rules, the corresponding reservation elements, real-time flight information fields, and seat inventory data fields are combined and coded to generate a structured reservation feature vector.

[0127] The internal structured coding rules include field unification rules, field coding rules, numerical normalization rules, context feature combination rules, field alignment rules, and coding output rules. These rules are used to standardize, vectorize, and combine the pre-defined elements, real-time flight information fields, and seat inventory data fields according to a preset field order and coding method, thereby generating a structured pre-defined feature vector.

[0128] In this embodiment, step three specifically includes:

[0129] Retrieves user's historical order data from the historical order database based on user identifier;

[0130] The historical order data includes at least the following fields: flight number, departure city, arrival city, travel date, departure time, arrival time, cabin class, actual price paid, carrier, and order status.

[0131] The historical order data is cleaned and processed to remove canceled orders, refunded orders, and incomplete order records. Missing fields are filled in, and the historical order data is sorted in chronological order.

[0132] Based on the sorted historical order data, the frequency of user selection on different flights is counted to generate a user's historical flight selection record;

[0133] Based on the sorted historical order data, the number of times and percentage of users selected each cabin class are statistically analyzed to generate a record of users' historical cabin class selections.

[0134] Based on the sorted historical order data, the number of orders and price distribution ranges of users in different price ranges are statistically analyzed to generate a record of users' historical price ranges.

[0135] Based on the sorted historical order data, the number and percentage of orders placed by users with different airlines are statistically analyzed to generate a record of users' historical airline selections.

[0136] Based on the sorted historical order data, the distribution of user departure times in different time intervals is statistically analyzed, and the departure times are mapped to preset time intervals to generate user historical time interval preference data.

[0137] Based on the user's historical price range records, the proportion of the user's selection in each price range is calculated to generate user price sensitivity characteristics that reflect the degree of price sensitivity.

[0138] The selection ratio of each time period is calculated based on users' historical time period preference data, and time period preference features reflecting the tendency to select time periods are generated.

[0139] The selection ratio of each cabin class is calculated based on the user's historical cabin class selection records, and cabin class preference features reflecting cabin class selection tendencies are generated.

[0140] The selection ratio of each airline is calculated based on the user's historical airline selection records, and airline preference features reflecting the airline selection tendency are generated.

[0141] User price sensitivity features, time period preference features, cabin class preference features and airline preference features are organized according to the preset field order and encoding method to form a user preference feature set, and the user preference feature set is normalized.

[0142] According to the internal user profile coding standard, the processed user preference feature set is mapped into a fixed-dimensional numerical vector to generate a user profile feature vector.

[0143] In this embodiment, step four specifically includes:

[0144] Read the fare, departure time, arrival time, cabin class, cabin availability and user behavior related features corresponding to the candidate flights from the structured booking feature vector, and read the feature values ​​corresponding to price sensitivity, time period preference, cabin class preference and airline preference from the user profile feature vector;

[0145] For each candidate flight, a price score, time matching score, user preference matching score, historical behavior similarity score, and cabin availability score are calculated based on the structured booking feature vector and user profile feature vector.

[0146] The time matching score is calculated based on the matching relationship between the departure and arrival times of candidate flights and the user's time preferences;

[0147] The preference matching score is calculated based on the matching relationship between the airline and cabin class of the candidate flight and the airline preference features and cabin class preference features in the user preference feature set;

[0148] Historical behavior similarity score is calculated based on the similarity between the candidate flight and the user's historical flight selection records;

[0149] The cabin availability score is calculated based on the relationship between the number of remaining seats and a preset inventory threshold.

[0150] Based on preset weights, the price score, time matching score, user preference matching score, historical behavior similarity score, and cabin availability score are weighted and calculated to obtain the comprehensive recommendation score for each candidate flight.

[0151] Semantic confidence scores corresponding to natural language ticket booking requests are obtained from semantic recognition processing, and the comprehensive recommendation score and semantic confidence scores are combined according to a preset joint decision rule.

[0152] The semantic confidence score is obtained by weighting or combining the semantic encoding confidence score generated by the semantic encoding module of the pre-trained language model with the intent recognition confidence score and context parsing confidence score generated by the large language model parsing module, according to the pre-set confidence evaluation rules of the semantic understanding engine.

[0153] The preset joint decision-making rules include: weighting the comprehensive recommendation score and semantic confidence according to preset weights; performing threshold filtering on the weighted results to remove candidate flights that are below the preset comprehensive threshold; sorting the candidate flights that meet the threshold conditions according to the weighted results, and selecting the target flight and target cabin class based on the sorting results.

[0154] The candidate flights are ranked, and the candidate flights and their corresponding cabin classes that meet the preset conditions in terms of both comprehensive recommendation score and semantic confidence are selected as the target flights and target cabin classes.

[0155] Based on the target flight and target cabin class, extract the flight number, departure date, departure time, departure airport, arrival airport, cabin class, and passenger-related fields from the structured booking feature vector;

[0156] According to the internal reservation instruction coding standard, the field information is organized into a unified format to generate a standardized ticket reservation request containing flight number, departure date, departure time, cabin class and passenger information.

[0157] In this embodiment, step five specifically includes:

[0158] Obtain standardized ticket booking requests generated for target flights and target cabin classes from the decision-making calculation process, and parse the standardized ticket booking requests into an internally unified data format;

[0159] Based on the target system identifier contained in the standardized flight booking request, determine whether the booking target system is the airline system or the global distribution system;

[0160] And select the corresponding protocol adaptation rule based on the target system identifier;

[0161] According to the protocol adaptation rules, multi-source data parsing and field mapping processing are performed on the internally unified data format. The internal field names are matched with the field names of the target system's reservation interface to obtain an intermediate data structure that is consistent with the fields of the target system's reservation interface.

[0162] Based on the intermediate data structure and the target system's protocol specifications, a reservation instruction format is generated:

[0163] When the target system uses text commands, it generates a text reservation command containing flight number, departure date, departure time, cabin class, and passenger information.

[0164] When the target system uses structured messaging, the structured reservation message is generated according to the target system's field specifications.

[0165] When the target system adopts the interface-defined format, the reservation request message is generated according to the parameter format and field order specified by the interface.

[0166] The generated reservation instruction format is encapsulated into a communication request message and sent to the corresponding airline system or global distribution system through a preset communication channel. During the sending process, the request time, target address and request identifier are recorded.

[0167] Receive reservation response data from airline systems or global distribution systems, perform parsing processing on the reservation response data according to the protocol specifications of the target system, extract the reservation result status, reservation record code and returned inventory information, and associate the parsed reservation response with the corresponding standardized ticket booking request;

[0168] During the process of sending reservation requests and receiving reservation responses, the communication status is monitored. If a reservation response is not received after a preset time, the current situation is determined to be a communication timeout.

[0169] The reservation request is resent according to the preset retry rules. If the reservation request is still unsuccessful after the number of retries exceeds the limit, the reservation request is marked as a communication failure and the reason for the failure is recorded.

[0170] During the parsing of reservation response data, the message structure, number of fields, and field types are detected. When a message format that does not conform to the target system protocol specification is detected, the current situation is judged as a message parsing error, the corresponding error type and error field information are recorded, and the associated reservation request is marked as parsing failure.

[0171] After parsing, the inventory information returned in the reservation response data is compared with the corresponding seat inventory data in the internal structured reservation feature vector. When an inconsistency is detected in the inventory quantity or available status, the current situation is determined to be an inventory inconsistency, an inventory inconsistency flag is generated, and the corresponding reservation request is included in the reservation failure handling process.

[0172] In this embodiment, step six specifically includes:

[0173] Based on the reservation result status field in the reservation response data, the reservation result is judged. When the reservation result status is successful, the reservation record code is extracted from the reservation response data and associated with the corresponding standardized flight reservation request.

[0174] Based on the associated standardized flight booking request and reservation record code, ticketing request data is generated in accordance with the ticketing interface specification. The ticketing request data includes at least the reservation record code, flight number, departure date, cabin class, and passenger information. The generated ticketing request data is then encapsulated into a ticketing request message.

[0175] The ticketing request message is sent to the corresponding airline system or global distribution system through a preset communication channel, and the request identifier, sending time and target system identifier of the ticketing request are recorded.

[0176] The ticketing response message is parsed according to the ticketing interface protocol to extract the ticketing result status, reservation record code and electronic ticket number, and the parsing result is matched with the corresponding ticketing request.

[0177] When the ticketing result status is successful, the parsed reservation record code and electronic ticket number are written into the corresponding order record, the order status field in the order record is updated to the ticketed status, and the ticketing completion time and ticketing channel information are recorded.

[0178] When the ticketing result status is failed or an inventory inconsistency flag, message structure error, or missing field is detected during the parsing of the ticketing response message, the current ticketing request will be marked as a ticketing failure, and the reason for failure will be written into the failure reason field in the order record.

[0179] The current ticketing request is marked as a ticketing failure time, and the booking failure rollback process is initiated, generating a rollback instruction according to the preset rules of the booking failure rollback process;

[0180] When the rollback rule requires cancellation of a reservation, a cancellation request message containing the reservation record code is constructed and sent to the corresponding airline system or global distribution system.

[0181] Receive the cancellation response message, parse the cancellation response, when the cancellation is successful, update the order status in the order record to the reservation failure status or the cancelled status, and record the cancellation time and cancellation instruction identifier; when the cancellation fails, mark the current cancellation operation as rollback failure and record the reason for the rollback failure.

[0182] The ticketing request message, ticketing response message, reservation record code, electronic ticket number, order status update result, and the triggering conditions, rollback instruction sending record, and rollback result of the reservation failure rollback process are written to the log record.

[0183] The written log records are associated with log identifier, reservation identifier, order identifier and flight identifier to generate reservation log entries containing time identifier, request identifier, flight identifier, order identifier and processing status fields;

[0184] The generated pre-defined log entries are stored in a pre-defined log storage medium, and a retrieval index is created based on the time identifier, request identifier, and flight identifier.

[0185] The semantic understanding engine in this invention combines the capabilities of a large language model parsing module and a pre-trained language model semantic encoding module to achieve in-depth understanding and structured transformation of users' natural language booking requests. The system first uses a pre-trained language model to perform semantic encoding on the input natural language, generating an intermediate semantic vector. Based on this semantic vector, it performs contextual semantic parsing to identify syntactic structures and semantic relationships. Subsequently, the large language model parsing module further infers the user's intent, generating a semantic confidence score. Combined with intent recognition, it extracts booking elements such as departure point, destination, travel time, number of passengers, and cabin class preference. For example, for a user inputting "Book me a flight to Beijing tomorrow afternoon," the system can automatically identify the departure point, destination, and time preference, converting them into structured feature parameters for subsequent decision-making.

[0186] The user profiling and strategy analysis module extracts and constructs user behavioral characteristics and booking preference information, forming a standardized user profile feature vector. Based on user identifiers, the system reads their historical order record database and performs statistical analysis on behavioral dimensions such as route, airline, cabin class, price range, and time preference. Using K-means clustering or collaborative filtering algorithms, it mines potential user preferences, extracting tag features such as price sensitivity, flight time preference, cabin class preference, and airline preference. This module normalizes the extracted features and outputs them as fixed-dimensional numerical vectors for subsequent multi-dimensional intelligent decision-making modules to achieve personalized booking strategy matching.

[0187] The multi-dimensional intelligent decision-making module constructs an intelligent scoring model based on structured booking feature vectors and user profile feature vectors, and outputs the optimal booking solution. This module utilizes a multi-objective weighted strategy to score candidate flights based on price matching, time preference matching, airline and cabin class matching, historical behavior similarity, and inventory availability. The scores for each dimension are weighted according to a preset weight matrix to obtain a comprehensive recommendation score, which is then combined with the semantic confidence score provided by the aforementioned semantic understanding module to perform flight ranking and filtering. Finally, a standardized flight booking request that matches user preferences, offers a reasonable price, and has sufficient inventory is output.

[0188] The unified data interface adaptation module is primarily responsible for converting standardized reservation requests generated by the system into the data format required by external airline systems or Global Distribution Systems (GDS). The module integrates multiple API adapters, supporting interfaces from mainstream airlines and GDS systems such as Eterm, Amadeus, China Southern Airlines NDC, and China Eastern Airlines NDC. The system uses a unified data model (UDM) to map parameters, standardize formats, and convert protocols, generating reservation messages that conform to the target interface. The module supports asynchronous communication mechanisms, capable of receiving and parsing PNR encoding, inventory status, and other information in reservation responses. It also possesses the ability to detect and handle anomalies such as communication timeouts, abnormal message structures, and inventory inconsistencies, ensuring the stability and controllability of the system's external interface communication.

[0189] After a successful reservation, the automatic ticketing module automatically calls the airline's or GDS ticketing interface to process the ticket. The module receives the PNR code from the reservation response, constructs standard ticketing request data, and sends it to the target ticketing interface. Upon successful ticketing, the system writes the Electronic Ticket Number (ETKT) and reservation record code into the order database and updates the order status to "Ticketed." If ticketing fails or anomalies are detected, such as inventory discrepancies, incorrect message structure, or missing fields, the system automatically triggers a reservation failure rollback process, cancels the reservation, and records the fault type and processing result. This module maintains real-time communication with the data interface adaptation module, achieving closed-loop management of the ticketing process.

[0190] The reservation log management module records and tracks key data throughout the system's operation. Each reservation and ticketing operation generates a unique log identifier ID, recording detailed log entries including user input data, intermediate AI parsing results, recommendation algorithm scoring details, API request and response content, exception handling records, and ticketing receipt information. The system builds a searchable index to enable full-process backtracking of any reservation, facilitating subsequent analysis, monitoring, and exception handling. This module also integrates an access control mechanism to ensure the security and compliance of log data in multi-user scenarios.

[0191] Example 1:

[0192] To verify the feasibility of this invention in practice, it was applied to the real-world business environment of an online ticketing platform. The entire process of a user booking an automated airline ticket using natural language was simulated, and the performance of the system in actual operation was observed. The platform previously used manual customer service to assist with bookings, which suffered from slow response times, low accuracy, and inefficient ticketing, resulting in a poor user experience. By introducing this invention, the platform aims to address practical pain points such as difficulties in understanding natural language, insufficient personalized recommendations, and untimely booking responses.

[0193] In the experimental scenario, the system integrated approximately 200,000 real historical ticket booking orders, covering multiple airlines, dozens of major routes, different price ranges, cabin classes, and departure times. By analyzing users' historical behavior, user profile feature vectors were generated, which were then used to match and recommend users against the results of natural language request parsing.

[0194] During the testing phase, 300 real registered users were randomly selected. Input was natural language phrases such as "Book me a flight to XX tomorrow afternoon," with users not required to provide complete structured information. The system extracted elements such as departure location, destination, and time preference using a semantic understanding engine, and generated semantic confidence scores using a large language model parsing module. The user profiling and strategy analysis module then calculated the probability of users' past choices in airline preference, cabin class, price range, and departure time, matching the optimal flight and seat.

[0195] The system uses a multi-dimensional intelligent decision-making module to calculate time matching score, airline preference matching degree, price score, and inventory tightness score, and combines these with semantic confidence to determine the final booking request. The system automatically completes the booking process and connects to China Southern Airlines NDC, China Eastern Airlines NDC, and a certain GDS system through a unified data interface adaptation module. After successful booking, the automatic ticketing module is triggered to generate an electronic ticket number. The entire process requires no manual intervention. Specific experimental data is shown in Table 1.

[0196] Table 1 Comparison of Actual Results of AI-Powered Automatic Ticket Booking Systems

[0197] Input text example System response time (seconds) Overall score of recommended solution Final ticketing status User satisfaction rating Does it trigger an exception process? Please book me a flight to XX tomorrow morning. 16.4 89.2 success 9.6 no Going to XX tomorrow afternoon, somewhere with a lower price. 18.1 84.3 success 9.0 no Fly to XX on Friday, the cheaper the better. 17.7 91.6 success 9.8 no Book a flight to XX after 6 PM. 21.2 85.1 success 8.9 yes Book a flight to XX for tomorrow afternoon, preferably in a comfortable cabin. 15.9 92.7 success 9.7 no I'd like to book a cheaper flight to XX tomorrow morning. 19.5 88.0 success 9.4 no I want to take a flight that departs before 10 o'clock to XX 17.3 86.2 success 9.1 no Please book me a ticket to XX for today, cheap is fine. 20.4 82.3 success 8.5 yes Flights departing for XX after 8 PM, economy class. 18.9 87.8 success 9.2 no I need a flight to XX starting around 8 AM tomorrow, as long as there are available seats. 16.8 90.1 success 9.5 no

[0198] The experimental results in Table 1 show that the system achieved an accuracy rate of 94.7% in one-time structured element extraction, an average semantic confidence score of 0.913, and an average comprehensive score of 87.2 for recommended solutions. The average time from natural language input to ticket issuance was controlled within 18 seconds, while the control group (human assistance) took approximately 2.5 minutes on average, representing an efficiency improvement of about 7 times. Particularly among price-sensitive users (whose price preferences are concentrated in the 400-800 yuan range), the system's recommendations highly matched the users' final purchase intentions, resulting in a 19.4% increase in conversion rate.

[0199] Regarding the system's anomaly handling capabilities, a total of 12 communication timeout cases, 7 inventory inconsistency cases, and 5 format message parsing anomalies were identified and handled. All of these were completed under the retry mechanism and rollback process, and no user complaints or system interruption issues occurred, demonstrating good stability.

[0200] The system of this invention not only achieves high-precision parsing of natural language intent, but also combines user historical behavior to complete personalized recommendations and intelligent decision-making, effectively improving ticketing processing efficiency, user satisfaction and system stability, fully demonstrating the technical advantages and business value of this invention in actual civil aviation automatic reservation scenarios.

[0201] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A civil aviation ticket automatic reservation system based on semantic recognition, characterized in that, include: The reservation input data acquisition module is used to acquire the natural language ticket booking request and related passenger data entered by the user through the terminal, and form reservation input data; The semantic understanding engine module is used to perform semantic recognition processing on natural language ticket booking requests and generate structured booking feature vectors; The large language model parsing module is used to perform semantic representation, contextual reasoning, and semantic confidence generation on natural language ticket booking requests. The pre-trained language model semantic encoding module is used to perform semantic encoding on natural language ticket booking requests and generate intermediate semantic vectors; The user profiling and strategy analysis module is used to perform statistics and feature extraction on historical flight selection records, cabin class selection records, price range records, airline selection records, and time period preference data based on users' historical order data, and generate user profile feature vectors. The multidimensional intelligent decision-making module is used to input the structured booking feature vector and the user profile feature vector into the comprehensive scoring model. Based on the time matching degree score, user preference matching score, historical behavior similarity score and cabin availability score, it generates a comprehensive recommendation score and performs joint decision-making with semantic confidence to determine standardized flight booking requests. The unified data interface adaptation module is used to perform multi-source data parsing, field mapping, and protocol conversion on standardized flight booking requests, send booking requests to the corresponding system, and handle abnormal situations. The automatic ticketing module is used to construct a ticketing request message after a successful reservation, send a ticketing request according to the ticketing interface protocol, parse the ticketing response, and write the reservation record code and electronic ticket number into the order record. When ticketing fails, it executes a reservation failure rollback process. The reservation log management module is used to log reservation input data, semantic recognition results, user profile feature vectors, decision results, request and response messages during the interface adaptation phase, and reservation and ticketing results.

2. The automatic civil aviation ticket booking system based on semantic recognition according to claim 1, characterized in that, The modules are connected in the following way: Step 1: Obtain the natural language booking request entered by the user through the terminal and related passenger data to form the booking input data; Step 2: Perform semantic recognition processing on the natural language booking requests in the pre-booking input data, match the booking elements with real-time flight information and seat inventory data, and perform field mapping and structuring processing to generate a structured booking feature vector; Step 3: Perform statistical analysis and feature extraction on historical order data to generate a user preference feature set, and combine the user preference feature set in a unified format to form a user profile feature vector; Step 4: Input the structured booking feature vector and user profile feature vector into the decision calculation process, calculate the comprehensive recommendation score, and make a joint decision with the semantic confidence obtained from semantic recognition to determine the standardized flight booking request; Step 5: Perform protocol adaptation processing on standardized flight booking requests, send booking requests to the target system, and handle exceptions such as communication timeouts, inventory inconsistencies, and message parsing errors. Step Six: After receiving the confirmation data that the reservation has been successfully made, proceed with the ticketing process; If ticketing fails or inventory discrepancies are detected, a booking failure rollback process is executed.

3. The automatic civil aviation ticket booking system based on semantic recognition according to claim 2, characterized in that, Step one specifically involves: Based on the natural language booking request, a flight query request is sent to the preset flight data service to obtain real-time flight information that matches the natural language booking request from the flight data service; In accordance with the internal reservation data format specifications, the fields of natural language ticketing requests, relevant passenger data, real-time flight information and seat inventory data were aligned and aggregated. Generate a corresponding reservation identifier and time identifier for each reservation service, and organize the aggregated data into a unified structure to form reservation input data; Establish a correspondence between log identifiers and predefined identifiers, and write the predefined input data into the log record.

4. The automatic civil aviation ticket booking system based on semantic recognition according to claim 2, characterized in that, Step two specifically involves: Read the text field corresponding to the natural language ticket booking request from the pre-defined input data, perform character normalization and segmentation on the text field, and obtain the natural language text to be parsed; Perform semantic encoding on natural language text to generate intermediate semantic vectors; Context parsing is performed based on intermediate semantic vectors to determine the semantic relationships and syntactic structure in natural language ticket booking requests, and the context parsing results are obtained. Based on the context parsing results, intent extraction is performed to identify booking elements related to ticket booking; According to the internal structured coding rules, the corresponding reservation elements, real-time flight information fields and seat inventory data fields are combined and coded to generate a structured reservation feature vector. The intermediate semantic vectors, context parsing results, pre-defined elements, and structured pre-defined feature vectors corresponding to the current pre-defined business generated during semantic recognition processing are written into the log records, and associated with the corresponding pre-defined identifiers and log identifiers.

5. The automatic civil aviation ticket booking system based on semantic recognition according to claim 2, characterized in that, Step three specifically involves: Retrieves user's historical order data from the historical order database based on user identifier; Based on the sorted historical order data, the frequency of user selection on different flights is counted to generate a user's historical flight selection record; Based on the sorted historical order data, the number of times and percentage of users selected each cabin class are statistically analyzed to generate a record of users' historical cabin class selections. Based on the sorted historical order data, the number of orders and price distribution ranges of users in different price ranges are statistically analyzed to generate a record of users' historical price ranges. Based on the sorted historical order data, the number and proportion of orders placed by users on different airlines are statistically analyzed to generate a record of users' historical airline selections. The distribution of users' departure times in different time periods is statistically analyzed, and the departure times are mapped to preset time periods to generate users' historical time period preference data. Based on the user's historical price range records, the proportion of the user's selection in each price range is calculated to generate the user's price sensitivity characteristics; Calculate the selection ratio of each time period interval based on users' historical time period preference data to generate time period preference features; Based on the user's historical cabin class selection records, the selection ratio of each cabin class is calculated to generate cabin class preference features; Calculate the selection ratio of each airline based on the user's historical airline selection records to generate airline preference features; User price sensitivity features, time period preference features, cabin class preference features and airline preference features are organized according to the preset field order and encoding method to form a user preference feature set, and the user preference feature set is normalized. The processed user preference feature set is mapped to a fixed-dimensional numerical vector to generate a user profile feature vector; Write the user's historical order data summary, user preference feature set, and user profile feature vector generated from the user preference feature set into the log record, and record the correspondence between user identifier and reservation identifier.

6. The automatic civil aviation ticket booking system based on semantic recognition according to claim 2, characterized in that, Step four specifically involves: A time matching score is calculated based on the matching relationship between the departure and arrival times of candidate flights and the user's time preferences. A preference matching score is calculated based on the matching relationship between the airline and cabin class of the candidate flights and the airline preference features and cabin class preference features in the user preference feature set. A historical behavior similarity score is calculated based on the similarity between the candidate flight and the user's historical flight selection records. A cabin space availability score is calculated based on the relationship between the number of remaining seats and a preset inventory threshold. Based on preset weights, the price score, time matching score, user preference matching score, historical behavior similarity score, and cabin availability score are weighted and calculated to obtain the comprehensive recommendation score for each candidate flight. The candidate flights are ranked, and the candidate flights and their corresponding cabin classes that meet the preset conditions in terms of both comprehensive recommendation score and semantic confidence are selected as the target flights and target cabin classes. Based on the target flight and target cabin class, extract the field information of the target flight from the structured booking feature vector; According to the internal booking instruction coding standard, the field information is organized into a unified format to generate a standardized flight booking request; The price score, time matching score, user preference matching score, historical behavior similarity score, cabin availability score, comprehensive recommendation score, semantic confidence score, target flight and target cabin information, and the generated standardized ticket booking request for each candidate flight are written into the log record.

7. The automatic civil aviation ticket booking system based on semantic recognition according to claim 2, characterized in that, Step five specifically involves: Based on standardized flight booking requests, the target booking system is determined to be either the airline's system or the global distribution system; And select the corresponding protocol adaptation rule based on the target system identifier; According to the protocol adaptation rules, the internally unified data format is parsed to obtain an intermediate data structure that is consistent with the reservation interface fields of the target system. Based on the intermediate data structure and the target system's protocol specifications, a reservation instruction format is generated. The generated reservation instruction format is encapsulated into a communication request message and sent to the corresponding airline system or global distribution system through a preset communication channel; During the process of sending reservation requests and receiving reservation responses, the communication status is monitored. If a reservation response is not received after a preset time, the current situation is determined to be a communication timeout. The reservation request is resent according to the preset retry rules. If the reservation request is still unsuccessful after the number of retries is exceeded, the reservation request is marked as a communication failure. During the parsing of reservation response data, the message structure, number of fields, and field types are detected. When a message format that does not conform to the target system protocol specification is detected, the current situation is judged as a message parsing error, the corresponding error type and error field information are recorded, and the associated reservation request is marked as parsing failure. After parsing, the inventory information returned in the reservation response data is compared with the corresponding seat inventory data in the structured reservation feature vector; When an inconsistency is detected in the inventory quantity or the availability of goods, the current situation is judged as an inventory inconsistency, an inventory inconsistency flag is generated, and the corresponding reservation request is included in the reservation failure handling process. The reservation instruction format, reservation request communication message, target system identifier, communication status, reservation response data, and exception judgment results such as communication timeout, inventory inconsistency, and message parsing error generated during the protocol adaptation phase are written into the log record.

8. The automatic civil aviation ticket booking system based on semantic recognition according to claim 2, characterized in that, Step six specifically involves: Based on the reservation result status field in the reservation response data, the reservation result is judged. When the reservation result status is successful, the reservation record code is extracted from the reservation response data. Based on the associated standardized flight booking request and reservation record code, generate ticketing request data according to the ticketing interface specification, and encapsulate the generated ticketing request data into a ticketing request message. The ticketing request message is sent to the corresponding airline system or global distribution system through a pre-set communication channel; The ticketing response message is parsed according to the ticketing interface protocol to extract the ticketing result status, reservation record code, and electronic ticket number; When the ticketing result status is successful, the parsed reservation record code and electronic ticket number are written into the corresponding order record, the order status field in the order record is updated to the ticketed status, and the ticketing completion time and ticketing channel information are recorded. When the ticketing result status is failed, or an inventory inconsistency flag, message structure error, or field missing is detected, the current ticketing request is marked as ticketing failure, and the reason for failure is written into the failure reason field in the order record. The current ticketing request is marked as a ticketing failure time, and the booking failure rollback process is initiated, generating a rollback instruction according to the preset rules of the booking failure rollback process; When the rollback rule requires cancellation of a reservation, a cancellation request message containing the reservation record code is constructed and sent to the corresponding airline system or global distribution system. When canceling a reservation fails, mark the current cancellation operation as a rollback failure and record the reason for the rollback failure; The ticketing request message, ticketing response message, reservation record code, electronic ticket number, order status update result, and the triggering conditions, rollback instruction sending record, and rollback result of the reservation failure rollback process are written to the log record.

Images