Air ticket recommendation method and device based on large language model, and electronic equipment

Abstract

The invention discloses an air ticket recommendation method and device based on a large language model and electronic equipment, and relates to the field of artificial intelligence and the field of aviation services. The method comprises the following steps: in response to an air ticket query request initiated by a user, obtaining a service scene prompt word, a session history and service information through an intelligent agent; obtaining preference information of a user and state information of each route; a basic template is generated according to the service scene cue word, knowledge information is extracted from the session history, the service information, the preference information of the user and the state information of each route, the basic template is filled with the knowledge information, and a target knowledge cue word is obtained; and processing the air ticket query request according to the target knowledge cue through the large language model, and generating at least one piece of recommended air ticket information for the user. The technical problems that in the prior art, the air ticket recommendation accuracy for the user is low, and the user experience feeling is poor are solved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and aviation services, and more specifically, to a method, device and electronic device for recommending air tickets based on a large language model. Background Art

[0002] With the rapid development of artificial intelligence technology, its application in various industries is becoming more and more extensive, and the field of civil aviation sales is no exception. The traditional civil aviation sales method mainly relies on ticket data for service. However, this method has gradually exposed its limitations and cannot meet the growing and diversified needs of modern passengers.

[0003] At present, the sales model of most airlines mainly focuses on providing consumers with basic ticket information, such as the specific time of flight take-off and landing, ticket prices and other basic elements, so as to provide users with purchase options. However, this traditional service method has an obvious defect, that is, it does not fully consider the diversified needs of users in the ticket purchase process. In fact, the user's ticket purchase decision is a complex process, which is affected by many factors, including the user's personal business behavior pattern (such as business trips or private travel), personal preference settings (such as window seats or aisle seats, whether there are special meal needs, etc.), and real-time changing environmental factors (such as weather conditions, traffic congestion, emergencies at the destination, etc.). These factors work together in the user's decision-making process and play a vital role. Unfortunately, the existing sales model does not effectively integrate the above key information into the service, resulting in users lacking sufficient information support when purchasing tickets, and it is difficult to get the choice that is closest to their actual needs. This situation not only seriously affects the user's ticket purchase experience, making them feel inconvenient and dissatisfied, but also weakens the ability of airlines to provide personalized and high-quality services, limiting their market competitiveness and further improvement of service levels.

[0004] In summary, facing the rapid development of artificial intelligence technology and the growing diversified demands in the field of civil aviation sales, the existing traditional sales model has become inadequate. This model is limited to the provision of basic ticketing information and fails to fully consider the diverse factors of users, resulting in a lack of comprehensive information support for users when making ticket purchase decisions. This not only affects the user's ticket purchase experience, but also leads to low accuracy in airline ticket recommendations for users.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] The present application provides a method, device and electronic device for air ticket recommendation based on a large language model, so as to at least solve the technical problem that the existing air ticket recommendation process only refers to the starting point, end point and price of the current route, resulting in low accuracy of air ticket recommendation for users and poor user experience.

[0007] According to one aspect of the present application, a method for recommending air tickets based on a large language model is provided, comprising: in response to an air ticket query request initiated by a user, obtaining business scenario prompt words, conversation history and business information through an intelligent agent, wherein the business scenario prompt words are used to characterize the business scenario information corresponding to the intelligent agent, the conversation history is the historical interaction information between the user and the software system, and the business information is the information related to aviation business obtained from application services other than the intelligent agent based on the air ticket query request and the business scenario information corresponding to the intelligent agent; obtaining the user's preference information and the status information of each route, wherein the preference information at least includes the user's habit information when handling aviation business and the user's behavioral preference information when selecting aviation flights; generating a basic template based on the business scenario prompt words, and extracting knowledge information from the conversation history, business information, the user's preference information and the status information of each route, and filling the knowledge information into the basic template to obtain a target knowledge prompt word; processing the air ticket query request based on the target knowledge prompt word through the large language model, and generating at least one recommended air ticket information for the user.

[0008] Optionally, obtaining the user's preference information and the status information of each route includes: determining the user's habit information when handling aviation business and the user's behavioral preference information when selecting aviation flights from the session history and the user's historical itinerary information by means of keyword recognition to obtain the user's preference information; determining the fare information, flight punctuality rate, flight delay risk information and weather information along the route for each route from the business information; and using the fare information, flight punctuality rate, flight delay risk information and weather information along the route for each route as the status information of each route.

[0009] Optionally, generating a basic template based on business scenario prompt words includes: obtaining a first category of prompt words, a second category of prompt words and a third category of prompt words, wherein the first category of prompt words are used to prompt the large language model to use a parsing method when parsing user intentions, the second category of prompt words are used to prompt the large language model to use a reasoning angle when processing target knowledge prompts, and the third category of prompt words are used to constrain the data format filled in the basic template; generating a basic template based on the first category of prompt words, the second category of prompt words, the third category of prompt words and the business scenario prompt words.

[0010] Optionally, knowledge information is extracted from the conversation history, business information, user preference information and the status information of each route, including: using the same aviation business perspective to extract corresponding knowledge information from each of the conversation history, business information, user preference information and the status information of each route; and / or using different aviation business perspectives to extract corresponding knowledge information from each of the conversation history, business information, user preference information and the status information of each route.

[0011] Optionally, after extracting knowledge information from conversation history, business information, user preference information and status information of each route, the air ticket recommendation method based on a large language model also includes: performing semantic splicing processing on at least two extracted knowledge information according to semantic content corresponding to the knowledge information to obtain new knowledge information.

[0012] Optionally, knowledge information is filled into the basic template to obtain target knowledge prompts, including: based on the business scenario prompts in the basic template, information screening is performed on knowledge information extracted from session history, business information, user preference information, and status information of each route to obtain target content having a correlation coefficient with the business scenario prompts greater than a preset threshold; and the target content is filled into the basic template to obtain target knowledge prompts.

[0013] Optionally, the knowledge information is filled into the basic template to obtain a target knowledge prompt, including: based on the business scenario prompt words in the basic template, the knowledge information extracted from the business information is screened to obtain a first content in the knowledge information whose correlation coefficient with the business scenario prompt words is greater than a preset threshold; based on the basic template, the business scenario prompt words and the first content are integrated into a second content, and based on the second content, the knowledge information extracted from the session history is screened to obtain a third content in the knowledge information whose correlation coefficient with the second content is greater than a preset threshold; based on the basic template, the second content and the third content are integrated into a fourth content, and based on the fourth content, the knowledge information extracted from the user's preference information and the status information of each route is screened to obtain a fifth content in the knowledge information whose correlation coefficient with the fourth content is greater than a preset threshold; the fourth content and the fifth content are integrated and filled into the basic template to obtain the target knowledge prompt.

[0014] Optionally, the air ticket query request is processed according to the target knowledge prompt through the large language model to generate at least one recommended air ticket information for the user, including: performing intent parsing and air ticket information inference on the target knowledge prompt and the air ticket query request according to the first category prompt words and the second category prompt words through the large language model, and determining at least one recommended air ticket information that meets the user intent based on the predicted user intent and the air ticket information inference result, wherein the expression form of the recommended air ticket information includes at least one of text form, voice form, image form and video form.

[0015] Optionally, the air ticket recommendation method based on the large language model also includes: when the recommended air ticket information is expressed in voice form, determining the language, tone and sound style used when conveying the recommended air ticket information to the user through voice based on the target knowledge prompt using the large language model; when the recommended air ticket information is expressed in text form, determining the character form and text expression style used when conveying the recommended air ticket information to the user through text based on the target knowledge prompt using the large language model; when the recommended air ticket information is expressed in image form, determining the image elements and combination of image elements used when conveying the recommended air ticket information to the user through image based on the target knowledge prompt using the large language model; when the recommended air ticket information is expressed in image form, determining the audio elements and video elements used when conveying the recommended air ticket information to the user through image based on the target knowledge prompt using the large language model.

[0016] According to another aspect of the present application, a flight ticket recommendation device based on a large language model is also provided, which includes: a first acquisition unit, which is used to respond to a flight ticket query request initiated by a user, and obtain business scenario prompt words, conversation history and business information through an intelligent agent, wherein the business scenario prompt words are used to characterize the business scenario information corresponding to the intelligent agent, the conversation history is the historical interaction information between the user and the software system, and the business information is the information related to the aviation business obtained from application services other than the intelligent agent based on the flight ticket query request and the business scenario information corresponding to the intelligent agent; a second acquisition unit, which is used to obtain the user's preference information and the status information of each route, wherein the preference information at least includes the user's habit information when handling aviation business and the user's behavior preference information when selecting aviation flights; a first processing unit, which is used to generate a basic template based on the business scenario prompt words, and extract knowledge information from the conversation history, business information, the user's preference information and the status information of each route, and fill the knowledge information into the basic template to obtain a target knowledge prompt word; a second processing unit, which is used to process the flight ticket query request based on the target knowledge prompt word through a large language model to generate at least one recommended flight ticket information for the user.

[0017] According to another aspect of the present application, a computer-readable storage medium is also provided, wherein the computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned air ticket recommendation method based on the large language model.

[0018] According to another aspect of the present application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the above-mentioned air ticket recommendation method based on a large language model is executed when the program is running.

[0019] According to another aspect of the present application, a computer program product is also provided, comprising computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the above-mentioned air ticket recommendation method based on a large language model are implemented.

[0020] In the present application, firstly, in response to the air ticket query request initiated by the user, the business scenario prompt words, conversation history and business information are obtained through the intelligent agent, wherein the business scenario prompt words are used to characterize the business scenario information corresponding to the intelligent agent, the conversation history is the historical interaction information between the user and the software system, and the business information is the information related to the aviation business obtained from the application service other than the intelligent agent according to the air ticket query request and the business scenario information corresponding to the intelligent agent. Then, the user's preference information and the status information of each route are obtained, wherein the preference information at least includes the user's habit information when handling aviation business and the user's behavior preference information when selecting aviation flights. Subsequently, a basic template is generated based on the business scenario prompt words, and knowledge information is extracted from the conversation history, business information, user's preference information and the status information of each route, and the knowledge information is filled into the basic template to obtain the target knowledge prompt words. Finally, the air ticket query request is processed according to the target knowledge prompt words through the large language model to generate at least one recommended air ticket information for the user.

[0021] From the above content, it can be seen that, first of all, through the comprehensive application of business scenario prompts, conversation history and business information, the intelligent agent can deeply understand the specific needs and preferences of users, rather than just recommending based on the departure point, destination and price. For example, users may prefer to choose flights of specific airlines, prefer specific departure or arrival times, have special meal needs and other habitual information, as well as weather preferences, attention to flight punctuality and other behavioral preference information. Through the processing of large models, the intelligent agent can accurately capture and apply these preferences, significantly improving the level of personalization of air ticket recommendations. Secondly, in this application, the intelligent agent not only pays attention to user preferences, but also obtains the status information of each route in real time, including but not limited to flight punctuality, aircraft model information, cabin status, fare fluctuations and emergency warnings. This means that when recommending air tickets, the system can comprehensively consider multiple factors, provide more comprehensive and accurate flight information, and help users make more reasonable itinerary choices.

[0022] Furthermore, in this application, business scenario prompts are used to generate basic templates. The intelligent agent then combines conversation history, business information, user preferences and route status to extract key knowledge information and fill it into the template to form a target knowledge prompt. This design ensures that the big model can focus on using this knowledge information when processing air ticket query requests, optimize the reasoning process, and provide recommendation results that are closer to user needs. Finally, through the deep understanding and reasoning of the big model, the intelligent agent can optimize the processing of air ticket query requests. The generated recommendation information is not only based on factual data, but also integrates intelligent analysis, and can provide the best air ticket recommendation plan from multiple dimensions such as user historical behavior, current environmental conditions, and business needs.

[0023] It can be seen that this application has significantly improved the accuracy and personalization of air ticket recommendations through the application of the above-mentioned technical features, so that users can obtain recommendation results that are both in line with actual needs and highly relevant when querying flight information. This not only simplifies the user's decision-making process, but also significantly improves user satisfaction and experience, providing users with a one-stop intelligent civil aviation sales service, making it more convenient and comfortable for users to plan their travel. The application of the technical features of this application solves the problem of low accuracy and poor user experience caused by the fact that air ticket recommendations in the prior art only consider basic elements. By deeply mining user preferences, integrating route status information, and applying large model intelligence, it realizes the personalized and intelligent upgrade of air ticket recommendation services, greatly improving the user's ticket purchasing experience and satisfaction in the field of civil aviation sales. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0025] Figure 1 A hardware structure block diagram of a computer terminal for implementing a ticket recommendation method based on a large language model is shown;

[0026] Figure 2 is a flowchart of an optional air ticket recommendation method based on a large language model according to an embodiment of the present application;

[0027] Figure 3 is a schematic diagram of an optional intelligent agent working process according to an embodiment of the present application;

[0028] Figure 4 It is a schematic diagram of an optional air ticket recommendation device based on a large language model according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] It should also be noted that the collected information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data are in compliance with relevant laws, regulations and standards, necessary confidentiality measures are taken, and public order and good customs are not violated, and corresponding operation entrances are provided for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions, providing users with corresponding operation entrances for users to choose to agree or refuse the results of automated decision-making; if the user chooses to refuse, the expert decision-making process will be entered.

[0032] According to an embodiment of the present application, an embodiment of a method for recommending air tickets based on a large language model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0033] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing the air ticket recommendation method based on a large language model according to an embodiment of the present application is shown. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0034] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuit may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As in the large language model-based air ticket recommendation method involved in the embodiments of the present application, the data processing circuit serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0035] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the air ticket recommendation method based on the large language model in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned air ticket recommendation method based on the large language model is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0036] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0037] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0038] Under the above operating environment, this application provides Figure 2 The air ticket recommendation method based on the large language model is shown. Figure 2 FIG. 1 is a flow chart of an optional method for recommending air tickets based on a large language model according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0039] Step S201, in response to a flight ticket query request initiated by a user, obtain business scenario prompt words, conversation history and business information through an intelligent agent.

[0040] The business scenario prompt words are used to characterize the business scenario information corresponding to the intelligent agent. The session history is the historical interaction information between the user and the software system. The business information is the information related to the aviation business obtained from application services other than the intelligent agent based on the ticket query request and the business scenario information corresponding to the intelligent agent.

[0041] Optionally, business scenario prompt words are key information carriers when the agent interacts with the large language model, and are used to clarify the agent's current business environment and task requirements. In this application, for air ticket query requests, business scenario prompt words specifically represent the business scenario information of aviation sales, such as the destination, departure time, cabin preference, airline preference and other key elements of the query. Through these prompt words, the large language model can quickly understand the context of the request and the business goals of the agent, so that when processing the request, it can more specifically call and integrate relevant knowledge information to provide users with more personalized and accurate flight recommendations.

[0042] Secondly, the conversation history refers to all the interaction records between the user and the software system before the current conversation. In the context of this application, the conversation history includes but is not limited to the historical records of the user's ticket query, reservation, change of ticket and other operations in the civil aviation sales software, as well as the specific needs and preferences expressed by the user in the conversation. By analyzing the conversation history, the agent can identify the user's past behavior patterns, such as preferred airlines, frequent flight routes, accustomed fare ranges, whether they often fly with pets, etc. This information is crucial for building user portraits, predicting user preferences, and providing personalized services.

[0043] In addition, business information refers to information related to aviation business provided by other application services besides the agent itself. Business information covers various external data required for aviation sales, such as real-time flight schedules, fare information, airline policies, weather forecasts, traffic conditions, flight punctuality statistics, seasonal fare change trends, discount information, and cabin information for specific routes. The acquisition and integration of business information is the basis for the agent to provide users with accurate and comprehensive services, ensuring that the agent can obtain the latest and most complete aviation business data when processing ticket query requests, thereby generating the best flight recommendations for users.

[0044] It should be noted that an agent refers to an automated component that performs specific tasks in the system. It can autonomously interact with users and process business requests according to preset rules and logic. In the civil aviation sales scenario of this application, the agent is designed as a business expert agent, which is responsible for ticket query and recommendation services. It generates personalized and intelligent ticket recommendation suggestions through deep reasoning of a large language model based on business scenario prompts, conversation history and business information, as well as user preferences and route status.

[0045] Step S202, obtaining the user's preference information and the status information of each route.

[0046] In step S202, the preference information includes at least the user's habit information when handling aviation services and the user's behavior preference information when selecting aviation flights.

[0047] Optionally, preference information refers to the personal inclination or choice preference expressed or implied by the user when conducting aviation business, including the user's habit information when handling aviation business and the behavioral preference information when choosing aviation flights. Among them, the acquisition and application of preference information is crucial to enhancing the personalization and intelligence of civil aviation sales services.

[0048] Optionally, the user's habit information when handling aviation business refers to the fixed patterns or preferences formed by the user when using civil aviation sales software, such as the user's tendency to choose a specific airline, preference for morning or evening flights, frequent selection of business class or economy class, whether there are special meal requirements (vegetarian, gluten-free, etc.), whether wheelchair services are needed, and the user's preference for the trade-off between ticket prices and services, etc. These habit information are identified and integrated by the intelligent agent through historical session records and business operation history data to build a more accurate user portrait.

[0049] Optionally, the user's behavioral preference information when choosing an air flight involves the user's selection tendency in the flight decision-making process. For example, the user may be more inclined to choose a route with a shorter connecting time, prefer a direct flight to a connecting flight, have higher requirements for flight punctuality, or have a special preference for entertainment facilities during the flight. By analyzing the user's query pattern and selection behavior, the intelligent agent can gain insight into the user's preferences when choosing an air flight, and thus make more accurate judgments in air ticket recommendations.

[0050] In addition, route status information refers to all dynamic information related to a specific route. Route status information reflects the latest status of the route at the time of query, which is crucial for users to make ticket purchase decisions. In this application, the route status information obtained by the agent should include at least one of the following information:

[0051] Real-time flight schedule: This contains the estimated departure and arrival times of flights and is important for planning your trip.

[0052] Fare information: Details of fares for different classes of service and whether there are any special promotions or offers.

[0053] Flight punctuality: Through statistical analysis, it provides the average flight delay time or the probability of on-time departure, helping users evaluate the reliability of flights.

[0054] Weather forecast: Checking the weather conditions at your flight's origin and destination is essential for weather-sensitive travel arrangements.

[0055] Traffic conditions: Understand the traffic congestion near the airport to help users plan their arrival time at the airport.

[0056] Class information: Understand the remaining seats in different classes of the flight so as to provide seat options that better meet user needs.

[0057] Service policy: including luggage carrying regulations, pet policy, special meal provision, etc., to meet the specific needs of users.

[0058] Disaster warning: When a specific route may be affected by a natural disaster (such as a typhoon, volcanic eruption, etc.), the system should provide warning information to help users avoid risks.

[0059] By obtaining preference information and route status information, the intelligent agent can provide users with more personalized and comprehensive ticket query and recommendation services. Preference information enables the system to understand the user's specific needs and preferences, while route status information ensures that the recommended flight information is the latest and most accurate, thereby improving service satisfaction and user experience. Through the intelligent reasoning of the large model, the intelligent agent can comprehensively analyze this information and generate the best ticket recommendation for the user, effectively solving the problem of over-generalization of ticket recommendations and poor user experience in the existing technology.

[0060] Step S203, generating a basic template based on the business scenario prompt words, extracting knowledge information from the session history, business information, user preference information and the status information of each route, and filling the knowledge information into the basic template to obtain the target knowledge prompt words.

[0061] Optionally, business scenario prompts are the starting point for the interaction between the agent and the big model. Business scenario prompts accurately describe the current business scenario and requirements. In the scenario of air ticket query, business scenario prompts may include specific prompts such as "query international flights", "family travel", "carrying pets", "business travelers", etc. These prompts guide the big model to understand the specific business environment and goals of the agent, so as to generate relevant knowledge information more accurately.

[0062] The basic template is a structured framework generated under the guidance of the business scenario prompt words, which is used to fill and integrate subsequent knowledge information. The design of the basic template should be closely related to the business scenario prompt words. For example, in the air ticket query scenario, the template may include fields such as flight information, price information, route preferences, special service requirements, etc. These fields will be filled in the subsequent steps to meet the specific needs of users.

[0063] Optionally, knowledge information is useful information obtained by the agent in this application based on comprehensive analysis of business scenario prompt words, conversation history, business information, user preferences and route status information. The extraction and integration of knowledge information is an important part of the work of the agent. It includes the user's personalized needs (such as preference for direct flights rather than transfers, sensitivity to ticket prices, special service needs), real-time route status (such as flight punctuality, weather conditions, seat conditions), and historical behavior patterns (such as frequent flight routes, price range preferences), etc. The accuracy and comprehensiveness of knowledge information directly determines the quality of air ticket recommendations.

[0064] Optionally, the target knowledge prompt is an instruction statement generated by filling the extracted knowledge information into the basic template to guide the intelligent reasoning of the big model. This prompt integrates multiple information such as business scenarios, user preferences, and route status, so that the big model can fully consider these factors when processing air ticket query requests and generate more personalized and user-friendly air ticket recommendation information. The generation of target knowledge prompts is the core of the interaction between the intelligent agent and the big model. The target knowledge prompt not only contains specific business data, but also implies the intelligent agent's understanding of user needs and judgment of business scenarios. It is the key to intelligent services.

[0065] For example, when the user is a frequent business traveler who prefers direct flights and has high requirements for flight punctuality, the agent will identify the user's business traveler identity from the conversation history, find out the user's preference for direct flights from the user's query records, and obtain the punctuality data of the current route through business information. At the same time, the agent will extract weather forecasts and traffic conditions from the route status information to ensure that the recommended flights are not affected by bad weather or traffic. Based on this information, the basic template generated by the agent may include fields such as "business traveler preference", "direct flights", "high punctuality requirements", and "no bad weather interference". Subsequently, the agent fills this knowledge information into the template and generates target knowledge prompts, such as: "Please recommend direct flights for business travelers, give priority to flights with high punctuality, and ensure that flights are not affected by bad weather. At the same time, provide flight fare information and optional cabins."

[0066] Through this process, the target knowledge prompts received by the large model contain a comprehensive description of the user's needs, and can realize intelligent reasoning based on the user's specific preferences and the current status of the route, and finally generate personalized air ticket recommendations for the business traveler. This process fully reflects the important value of this application in improving the personalization and intelligence of services.

[0067] Step S204: Process the air ticket query request according to the target knowledge prompt through the large language model to generate at least one recommended air ticket information for the user.

[0068] Optionally, the large language model is a complex, deep learning-based artificial intelligence model that is trained with a large amount of text data and can understand and generate high-quality natural language text. In this application, the large language model is used as a core reasoning component, responsible for parsing and understanding the contextual information in the target knowledge prompt, including user preferences, route status, business scenario description, etc., and then intelligently reasoning based on this information to generate air ticket recommendation information that better meets user needs. The power of the large language model lies in its ability to capture the nuances of language and perform complex logical reasoning based on this information, such as understanding users' preferences for direct flights, evaluating the impact of weather conditions on flight selection, etc., thereby providing users with more accurate and humane services.

[0069] Recommended flight information is a flight option for specific user needs generated by the agent after processing the target knowledge prompt through a large language model. This is not just a simple list of prices and times, but contains comprehensive information about the user's personalized needs, such as flight punctuality, cabin service, price, airline reviews, and descriptions of the impact of weather and traffic conditions on flights. The generation of recommended flight information takes into account the specific preferences expressed by the user in the conversation history, the real-time needs for flight status, and the knowledge information extracted based on business scenario prompts and route status information. The agent ensures that each recommended flight information can meet the user's specific needs in the current scenario and provide the best travel options.

[0070] For example, when processing a user's flight query request, the target knowledge prompt may be described as follows: "The user prefers morning flights, has high requirements for flight punctuality, and hopes to have extra legroom." After receiving this prompt, the large language model will search for eligible flights in the database based on user preferences and route status information. It will give priority to flights that take off in the morning and have a high punctuality rate, and filter out cabins that provide extra legroom services. Ultimately, the large language model will generate one or more recommended flight information, each of which will detail the flight time, punctuality, cabin details (including the option of extra legroom), price, and airline policies, ensuring that users can make decisions that best suit their needs based on this information.

[0071] Through this technical feature, the intelligent agent of this application can generate highly personalized air ticket recommendation information with the help of the intelligent reasoning ability of the large language model, thereby significantly improving the user experience and satisfaction of civil aviation sales services. This reasoning mechanism based on knowledge prompts can not only accurately capture the personalized needs of users, but also flexibly respond to the ever-changing route status, making the air ticket recommendation service more intelligent, efficient and humane.

[0072] Based on the contents of the above steps S201 to S204, it can be seen that, firstly, through the comprehensive application of business scenario prompt words, conversation history and business information, the intelligent agent can deeply understand the specific needs and preferences of users, rather than just recommending based on the departure point, destination and price. For example, users may prefer to choose flights of specific airlines, prefer specific departure or arrival times, have habitual information such as special meal needs, as well as behavioral preference information such as weather preferences and attention to flight punctuality. Through the processing of large models, the intelligent agent can accurately capture and apply these preferences, significantly improving the personalization level of air ticket recommendations. Secondly, in this application, the intelligent agent not only pays attention to user preferences, but also obtains the status information of each route in real time, including but not limited to flight punctuality, aircraft model information, cabin status, fare fluctuations and emergency warnings. This means that when recommending air tickets, the system can comprehensively consider multiple factors, provide more comprehensive and accurate flight information, and help users make more reasonable itinerary choices.

[0073] Furthermore, in this application, business scenario prompts are used to generate basic templates. The intelligent agent then combines conversation history, business information, user preferences and route status to extract key knowledge information and fill it into the template to form a target knowledge prompt. This design ensures that the big model can focus on using this knowledge information when processing air ticket query requests, optimize the reasoning process, and provide recommendation results that are closer to user needs. Finally, through the deep understanding and reasoning of the big model, the intelligent agent can optimize the processing of air ticket query requests. The generated recommendation information is not only based on factual data, but also integrates intelligent analysis, and can provide the best air ticket recommendation plan from multiple dimensions such as user historical behavior, current environmental conditions, and business needs.

[0074] It can be seen that this application has significantly improved the accuracy and personalization of air ticket recommendations through the application of the above-mentioned technical features, so that users can obtain recommendation results that are both in line with actual needs and highly relevant when querying flight information. This not only simplifies the user's decision-making process, but also significantly improves user satisfaction and experience, providing users with a one-stop intelligent civil aviation sales service, making it more convenient and comfortable for users to plan their travel. The application of the technical features of this application solves the problem of low accuracy and poor user experience caused by the fact that air ticket recommendations in the prior art only consider basic elements. By deeply mining user preferences, integrating route status information, and applying large model intelligence, it realizes the personalized and intelligent upgrade of air ticket recommendation services, greatly improving the user's ticket purchasing experience and satisfaction in the field of civil aviation sales.

[0075] In an optional embodiment, in order to obtain the user's preference information and the status information of each route, the agent can determine the user's habit information when handling aviation business and the user's behavior preference information when choosing aviation flights from the conversation history and the user's historical itinerary information through keyword recognition, and obtain the user's preference information; determine the fare information, flight punctuality rate, flight delay risk information and weather information along the route for each route from the business information. Finally, the agent uses the fare information, flight punctuality rate, flight delay risk information and weather information along the route for each route as the status information of each route.

[0076] Optionally, keyword recognition is the basic technology for the agent to analyze user session history and historical itinerary information to determine user habits and behavioral preference information. The agent identifies keywords related to flight inquiries, reservations, and preferences by analyzing the text or voice conversations input by the user. For example, if the user mentioned information such as "I usually choose business class", "I like to fly directly and avoid transfers", and "I need wheelchair service" in past conversations, the agent can use keyword recognition to extract the user's preference for cabin space, preference for flight type, and special needs in aviation services. This information together constitutes part of the user's preference information. The accuracy and efficiency of keyword recognition are the key to the intelligent recommendation capabilities of the entire system, and can ensure that the agent extracts the most relevant information points from historical data.

[0077] By obtaining user preferences through keyword recognition and obtaining route status in real time from external business information, the agent can generate highly personalized and intelligent ticket recommendation information in the processing of ticket query requests. This recommendation mechanism not only takes into account the user's personal preferences, but also fully considers the actual status of the flight, thereby greatly improving the user experience and the intelligence level of civil aviation sales services.

[0078] For example, the intelligent agent discovers through keyword recognition that the user prefers business class and attaches importance to flight punctuality. It then combines this information with the route status information obtained from the business information (such as a direct flight has business class seats available, with an punctuality rate of 98%, and clear weather along the way), and integrates this information into route status information, thereby generating a recommendation message: "Based on your preferences, we recommend a certain airline's business class seats at 8 pm next Tuesday. The flight has a high punctuality rate and good weather conditions along the way, which is very suitable for your itinerary needs." This recommendation method not only reflects the personalization of the service, but also ensures the real-time and accuracy of the information. It is a concrete manifestation of the technical features of the intelligent agent in this application in practical applications.

[0079] Through the application of the above-mentioned technical features, the intelligent agent can provide users with more considerate, accurate and timely air ticket recommendations than traditional civil aviation sales services, thereby improving the user's ticket purchasing experience.

[0080] In an optional embodiment, the agent can obtain the first category of prompt words, the second category of prompt words, and the third category of prompt words, wherein the first category of prompt words are used to prompt the large language model to use the parsing method when parsing the user's intention, the second category of prompt words are used to prompt the large language model to choose the reasoning angle when processing the target knowledge prompt words, and the third category of prompt words are used to constrain the data format filled in the basic template. Then, the agent generates a basic template based on the first category of prompt words, the second category of prompt words, the third category of prompt words, and the business scenario prompt words.

[0081] Optionally, the first type of prompt words are directive words or phrases used to guide the large model on how to understand the user's intention when the agent interacts with the large language model. These prompt words may include, but are not limited to, parsing user emotions, identifying the urgency of user needs, understanding the business type of user queries, etc. For example, when processing a user's flight query request, the first type of prompt words may be: "Please identify whether the user prefers morning flights", "Determine whether the user asks about the flight delay compensation policy", "Analyze whether the user is sensitive to ticket prices", etc. Through these prompts, the large model can more accurately understand the user's intentions and needs in a specific session, thereby generating response content that is more in line with user needs.

[0082] The second type of prompt words is to ensure that the large language model can reason and integrate information from the correct angle or dimension when processing the target knowledge prompt words. These prompt words are usually related to specific business scenarios, helping the large model focus on the most relevant information points and reasoning directions. For example, in the air ticket recommendation scenario, the second type of prompt words may include: "Optimize flight recommendations from a price perspective", "Consider flight punctuality for recommendation sorting", "Provide personalized recommendation reasons based on user behavior preferences", etc. These prompt words ensure that the large model can comprehensively consider multiple dimensions such as ticket prices, punctuality, user preferences, etc. when processing air ticket query requests, and provide users with the most reasonable and personalized recommendation solutions.

[0083] The third type of prompts is to standardize the data format when the large language model generates response information, ensuring the structure and readability of the information. This type of prompts can be to require the large model to present information in a specific structure (such as tables, lists, etc.), use specific terms or abbreviations, or follow a specific logical order. For example, if the target knowledge prompt generated by the agent requires the display of detailed information for each flight, the third type of prompts may be: "Please list the take-off and landing times, fares, and cabin information of all flights in a table", "Use airline codes instead of full names", etc. These format constraints make the response information generated by the large model clearer and more organized, which is convenient for the agent to process and display to the user.

[0084] Then, the agent generates a basic template according to the first category prompt words, the second category prompt words, the third category prompt words and the business scenario prompt words.

[0085] By using these three types of prompt words in combination, the agent can ensure that the large language model can accurately understand the user's needs when processing requests, and can start from the correct reasoning perspective, while ensuring the structured and standardized presentation of information, thereby providing users with efficient, accurate and easy-to-understand civil aviation sales services. Through this mechanism, the interaction between the agent and the large language model becomes more intelligent and customized, significantly improving the overall quality of the service and user experience.

[0086] In an optional embodiment, in order to extract knowledge information from the conversation history, business information, user preference information and the status information of each route, the intelligent agent may use the same aviation business perspective to extract corresponding knowledge information from each of the conversation history, business information, user preference information and the status information of each route; and / or use different aviation business perspectives to extract corresponding knowledge information from each of the conversation history, business information, user preference information and the status information of each route.

[0087] Optionally, the agent can extract knowledge information from the same aviation business perspective. When the agent extracts knowledge from information from different sources using the same aviation business perspective, it means that the agent will uniformly process and integrate data from session history, business information, user preferences, and route status around a specific business requirement or standard. For example, in a flight query scenario, the agent may choose "punctuality" as a unified business perspective, identify from the session history whether the user has ever expressed dissatisfaction with flight delays, obtain the on-time rate data of the current route from the business information, analyze from the user preference information whether the user tends to choose flights with high on-time rates, and monitor the actual on-time rate from the route status information. Through this perspective, the agent can synthesize all relevant information and generate knowledge information about flight on-time rates, such as "users prefer flights with high on-time rates, the average on-time rate of the current route is 95%, and the user has mentioned dissatisfaction with delayed flights in the historical session. It is recommended to give priority to flights with high on-time rates."

[0088] On the other hand, the agent can also extract knowledge from various information sources based on different aviation business perspectives to meet the needs of users in multiple dimensions. For example, in a flight query scenario, the agent may use "price", "service" and "convenience" as different business perspectives, identify whether the user is price sensitive from the conversation history, analyze whether the user prefers business class service from the user preference information, and obtain whether the flight's take-off and landing times meet the user's time window requirements from the route status information. In this way, the agent can generate multi-dimensional knowledge information, such as "the user prefers economy fares, business class services, and hopes that the flight will take off and land in the afternoon on weekdays", so that when recommending flights, it can fully consider the user's various needs and provide more personalized and accurate services.

[0089] Whether extracting knowledge information from the same aviation business perspective or from different business perspectives, the purpose of the agent is to better understand and meet the specific needs of users. By comprehensively acquiring and processing knowledge from multiple information sources, the agent can build a more comprehensive and in-depth user demand model, which not only helps to improve the accuracy and satisfaction of flight recommendations, but also enhances the intelligence and humanity of aviation services.

[0090] In an optional embodiment, after extracting knowledge information from conversation history, business information, user preference information and status information of each route, the intelligent agent can perform semantic splicing processing on at least two extracted knowledge information according to the semantic content corresponding to the knowledge information to obtain new knowledge information.

[0091] Alternatively, semantic content refers to the meaning and context in knowledge information, describing the intrinsic meaning of information and its association with context. In the airline sales scenario, the semantic content analyzed by the agent may include "user preference for business class", "high demand for flight punctuality", "user time preference for booking flights", etc. These semantic contents are not limited to superficial keywords, but also involve deeper understanding, such as the potential meaning in the user's words, implicit demand for services, etc.

[0092] In addition, semantic splicing processing refers to the intelligent agent logically combining multiple pieces of extracted knowledge information according to their semantic content to generate more comprehensive and insightful new knowledge information. This process involves the analysis, understanding and integration of knowledge information, aiming to create a comprehensive knowledge view that reflects user needs and scenario status. For example, the intelligent agent can splice the knowledge information of "user preference for business class" and "flight punctuality" to generate "information that users prefer business class and require high flight punctuality". In this way, the intelligent agent can generate new knowledge information that better meets user needs and scenario complexity, providing richer and more accurate basis for subsequent decision-making and recommendations.

[0093] Optionally, the new knowledge information is obtained through the above-mentioned semantic splicing process, and contains comprehensive information with more comprehensive and deeper meanings. The generation of this information not only improves the agent's ability to understand user needs, but also enables it to provide users with more personalized services based on more detailed and precise scene states. For example, a new knowledge information may be: "The user prefers business class, requires a flight punctuality rate of more than 90%, is price sensitive, and hopes to travel on weekdays to avoid peak hours." Such information not only includes the user's preferences for cabin space and punctuality, but also takes into account price factors and special needs of travel time. It is an important reference for the agent to recommend flights to users.

[0094] In an optional embodiment, knowledge information is filled into a basic template to obtain a target knowledge prompt, including: based on the business scenario prompt words in the basic template, information is filtered on knowledge information extracted from session history, business information, user preference information, and status information of each route to obtain target content having a correlation coefficient with the business scenario prompt words greater than a preset threshold; and the target content is filled into the basic template to obtain a target knowledge prompt.

[0095] Optionally, information screening refers to the evaluation and selection of knowledge information extracted from multiple information sources by the agent before generating the target knowledge prompt, ensuring that only content that is highly relevant to the business scenario prompt is retained. This process usually involves the calculation of a correlation coefficient, that is, evaluating the degree of association between the knowledge information and the business scenario prompt. Only when the correlation coefficient of the knowledge information exceeds a preset threshold will they be considered as target content and then filled into the basic template. Target content refers to the knowledge information retained after information screening, which is considered to be highly relevant to the user's intention and business scenario, and is the direct basis for the agent to generate the target knowledge prompt.

[0096] In an optional embodiment, the knowledge information is filled into the basic template to obtain the target knowledge prompt, including: first, the intelligent agent performs information screening on the knowledge information extracted from the business information according to the business scenario prompt in the basic template, and obtains the first content in the knowledge information whose correlation coefficient with the business scenario prompt is greater than the preset threshold. Then, the intelligent agent integrates the business scenario prompt and the first content into the second content according to the basic template, and performs information screening on the knowledge information extracted from the conversation history according to the second content, and obtains the third content in the knowledge information whose correlation coefficient with the second content is greater than the preset threshold. Subsequently, the intelligent agent integrates the second content and the third content into the fourth content according to the basic template, and performs information screening on the knowledge information extracted from the user's preference information and the status information of each route according to the fourth content, and obtains the fifth content in the knowledge information whose correlation coefficient with the fourth content is greater than the preset threshold. Finally, the intelligent agent integrates the fourth content and the fifth content and fills them into the basic template to obtain the target knowledge prompt.

[0097] Optionally, the present application provides an information screening and integration strategy using a "snowball" style, whereby the intelligent agent can achieve knowledge information integration from coarse to fine, from local to overall, significantly improving the accuracy and efficiency of information processing. First, the intelligent agent screens the knowledge information in the business information based on the business scenario prompt words, and the intelligent agent obtains the first content that is most directly related to the user's request. This preliminary screening ensures that the focus of subsequent processing is always on the core needs of the user. Secondly, information screening is not completed in one go, but is carried out in layers. First, the business scenario prompt words are integrated with the first content into the second content to provide more accurate guidance for further session history screening. Next, the second content and the third content are integrated, and then screened and integrated with the user's preference information and route status information to form the fourth content and the fifth content. This process is like a snowball, gradually increasing the richness and pertinence of the information until the target knowledge prompt is finally generated.

[0098] In addition, it should be noted that by hierarchically integrating user preference information and route status information, the agent can generate more personalized target knowledge prompts. This prompt not only takes into account the user's business needs, but also incorporates the user's personal preferences and real-time environmental information, providing users with a one-stop, comprehensive flight query and recommendation service. Using the "snowball" screening strategy, the agent can efficiently filter out the most relevant content from a large amount of information while ensuring the accuracy of the screening process. In each round of screening, the correlation coefficient is evaluated to avoid the interference of irrelevant information, ensuring that the target knowledge prompts generated in the end are both comprehensive and accurate. This strategy is particularly suitable for dealing with complex and changeable business scenarios, such as flight query and recommendation. By gradually screening and integrating knowledge information from different sources, the agent can adapt to the diversity of user needs, while considering the dynamic changes of business scenarios, and provide users with more flexible and detailed personalized services.

[0099] From the above content, it can be seen that the "snowball" information screening and integration strategy in this application ensures that the target knowledge prompts generated by the intelligent agent can accurately reflect user needs through gradual screening and precise integration. At the same time, considering the complexity of the business scenario, it greatly improves the service quality and personalization level of the intelligent agent in the field of aviation sales. The application of this strategy not only demonstrates the innovation of the intelligent agent in knowledge management and personalized services, but also reflects its efficiency and intelligence in handling complex business scenarios.

[0100] In an optional embodiment, the intelligent agent can use a large language model to perform intent analysis and air ticket information inference on the target knowledge prompt and the air ticket query request based on the first category prompt words and the second category prompt words, and determine at least one recommended air ticket information that meets the user's intent based on the predicted user intent and the air ticket information inference result, wherein the recommended air ticket information is expressed in at least one of text, voice, image, and video forms.

[0101] Optionally, when the recommended air ticket information is in voice form, the large language model is used to determine the language, tone and sound style used when conveying the recommended air ticket information to the user via voice based on the target knowledge prompt; when the recommended air ticket information is in text form, the large language model is used to determine the character form and text expression style used when conveying the recommended air ticket information to the user via text based on the target knowledge prompt; when the recommended air ticket information is in image form, the large language model is used to determine the image elements and combination of image elements used when conveying the recommended air ticket information to the user via image based on the target knowledge prompt; when the recommended air ticket information is in image form, the large language model is used to determine the audio elements and video elements used when conveying the recommended air ticket information to the user via image based on the target knowledge prompt.

[0102] Optionally, in the above content. Air ticket information reasoning refers to the agent using a large language model to infer the air ticket information that best meets the user's needs based on the target knowledge prompt and the user's query request, combined with business information, user preferences and real-time environmental information. This reasoning process is not just a simple data matching, but through the model's powerful reasoning ability and in-depth understanding of knowledge information, it provides users with more comprehensive and personalized air ticket recommendations. For example, the large language model may take into account the flight's punctuality history, special services (such as pet carrying, meal requirements), fare fluctuation trends, etc., to recommend a comprehensive cost-effective flight to the user.

[0103] Optionally, the presentation form of the recommended air ticket information refers to the way in which the agent conveys the recommendation result to the user, including at least one of text, voice, image, and video. The agent selects an appropriate presentation form based on user preferences and interaction scenarios to enhance the user experience. For example, for users who are accustomed to using voice assistants, the agent may choose voice; for visually oriented users, it may use images or videos to convey information.

[0104] The agent can use the large language model to determine personalized communication methods for recommended flight information in different forms. In voice form, the agent predicts the user's possible language, tone, and voice style preferences based on the target knowledge prompts to ensure that the voice communication is more in line with the user's preferences. In text form, the agent may adjust the wording of the text to use the user's preferred character form and expression style, such as formal or casual language style. In image or video form, the agent will design images with specific image elements and combinations based on the target knowledge prompts, or select appropriate audio and video elements to present the recommended information in a visual and auditory way.

[0105] When using voice to express flight recommendation information, the large language model can predict the language (Mandarin or English), tone (professional or friendly), and voice style (male or female voice) that the user may prefer based on the user's preferences (such as interest in business class) and business scenarios (such as flight punctuality) in the target knowledge prompts. The agent then uses these predictions to convey the recommendation results in the user's preferred voice, making the interaction process more humane.

[0106] When expressing flight recommendation information in text form, by analyzing the user preference information in the target knowledge prompt, the large language model can adjust the style of text expression, such as more formal or more informal wording, to match the user's reading preference. When presenting recommendation information, the agent can use the user's preferred character form (such as simplified Chinese or English), as well as the text style that combines the user's emotions and context to improve the attractiveness and readability of the text.

[0107] When using images to express flight recommendation information, the large language model can design recommended images containing specific image elements (such as pet logos and wheelchair icons) based on the special needs of user queries (such as bringing pets or requiring wheelchair services) through target knowledge prompts. By adjusting the combination of image elements, the agent can intuitively present the features of the recommended flights, such as highlighting business class services or special meal options in the image, so that users can get key information at a glance.

[0108] When using images to express flight recommendation information, the large language model can create recommendation content containing audio and video elements based on the target knowledge prompts. The audio element may include a voice description of the recommended flight, while the video element can show the internal facilities of the flight, the service process, or the introduction of the destination's attractions. Through carefully designed audio and video elements, the intelligent agent can convey the recommended flight information in a more vivid and attractive form, enhancing the user's visual and auditory experience.

[0109] Through the application of the above personalized communication methods, the intelligent agent can not only accurately understand user needs and business scenarios through the large language model, but also convey recommended air ticket information in the most appropriate form according to user preferences and interaction habits, thereby significantly improving the personalization level of services and user experience. The application of this innovative technology provides new directions and development space for intelligent services in the field of aviation sales.

[0110] In an optional embodiment, Figure 3 is a schematic diagram of an optional agent working process according to an embodiment of the present application, such as Figure 3 As shown in the figure, the composition of the business expert agent includes: business scenario prompt words, conversation history (corresponding to Figure 3session records in the database), business information (corresponding to Figure 3 business data generated by business services in the business), memory module, knowledge application prompt words, prompt word synthesizer and knowledge wisdom application prompt words.

[0111] Alternatively, if Figure 3 As shown, the memory module contains a knowledge set component, which stores three key knowledge: behavioral preference knowledge, business knowledge, and environmental knowledge.

[0112] Business scenario prompts are written based on the current business needs of the intelligent agent, aiming to guide the big model to parse user intent and construct business service request objects; at the same time, when processing business scenario responses, they can also guide the big model to perform intelligent reasoning, optimize the response content from multiple angles, and ensure its accuracy and effectiveness. The session history records all request and response interactions between the user and the system. Business information is relevant business information obtained from external dependent services based on user requests and current business context.

[0113] Optionally, the memory module is responsible for building a knowledge set by integrating information from multiple raw data sources, such as past conversation records between users and business systems, business operation history of users in business systems, and real-time environmental information. In particular, environmental information (such as Figure 3 As shown in the figure, it can be obtained from weather services, traffic services, and disaster warning services. It is closely related to factors such as time (such as policy restrictions) and geographical location (such as local climate and traffic conditions). Knowledge acquisition capabilities can be integrated in the knowledge construction stage, and dynamic information integration can be performed in the knowledge application stage. By analyzing this information, the user's personal preference information and taboos, business habits and behaviors, and other related business information, weather conditions, traffic information, and possible accident warnings can be extracted, and these contents can be converted into knowledge that is helpful for specific business scenarios, namely behavioral preference knowledge, business knowledge, and environmental knowledge.

[0114] Knowledge application prompts are written around different categories and business perspectives of the extracted knowledge set to guide the large model to make full use of this knowledge in reasoning and optimize the details, thereby improving the pertinence and effectiveness of reasoning. For the breadth expansion of the business, knowledge application prompts need to be written based on different business perspectives, and for the depth expansion of the business, knowledge application prompts need to be written based on the same business perspective.

[0115] The prompt word synthesizer is mainly used to synthesize prompt words for the intelligent application of knowledge when processing business scenario responses. It uses business scenario prompt words as the basic template, combines multiple knowledge application prompt words from different business perspectives, and is assisted by user conversations, business information, and knowledge sets to synthesize multiple knowledge application prompt words.

[0116] Knowledge and wisdom application prompts are used to guide large models to optimize business information from different business perspectives.

[0117] The knowledge set component is responsible for storing the knowledge extracted through the memory module and participating in the synthesis of prompt words when the knowledge set is used intelligently.

[0118] An optional service flow is as follows:

[0119] Step 1: The business expert agent receives the request information raised by the user in the conversation.

[0120] Step 2: Based on the business scenario prompt words and user conversation content, the intelligent agent requests the big model service to resolve the business services that need to be accessed.

[0121] Step 3: The big model analyzes the business scenario prompt words and user session content, determines the business services that need to be accessed and the necessary service request objects, and returns them to the agent in the form of structured data.

[0122] Step 4: The agent uses these structured data to request corresponding business services.

[0123] Step 5: The business service returns relevant data based on the request content, and the agent processes the data and stores it in the business information module.

[0124] Step 6: Through the prompt word synthesizer, the intelligent agent integrates user requests, business information, business scenario prompt words, knowledge set content and knowledge application prompt words to generate multiple knowledge intelligence application prompt words for different business perspectives.

[0125] Step 7: The business expert agent passes these prompt words to the large model service in multiple rounds to optimize the user request response. For optimization tasks with dependencies, serial processing is used, while for those without dependencies, parallel processing is used to improve efficiency.

[0126] Step 8: The intelligent agent organizes and merges the user request responses that have undergone multiple rounds of optimization and assembles them into the best response that is ultimately suitable for returning to the user.

[0127] Step 9: The business application returns the final response to the user, who can then initiate a new request.

[0128] From the above content, it can be seen that through the comprehensive application of business scenario prompts, conversation history and business information, the intelligent agent can deeply understand the specific needs and preferences of users, rather than just recommending based on the departure point, destination and price. For example, users may prefer to choose flights of specific airlines, prefer specific departure or arrival times, have special meal needs and other habitual information, as well as weather preferences, attention to flight punctuality and other behavioral preference information. Through the processing of large models, the intelligent agent can accurately capture and apply these preferences, significantly improving the level of personalization of air ticket recommendations. Secondly, in this application, the intelligent agent not only pays attention to user preferences, but also obtains the status information of each route in real time, including but not limited to flight punctuality, aircraft model information, cabin status, fare fluctuations and emergency warnings. This means that when recommending air tickets, the system can comprehensively consider multiple factors, provide more comprehensive and accurate flight information, and help users make more reasonable itinerary choices.

[0129] According to another aspect of the present application, a device for recommending air tickets based on a large language model is also provided, wherein: Figure 4 is a schematic diagram of an optional air ticket recommendation device based on a large language model according to an embodiment of the present application, such as Figure 4 As shown, it includes: a first acquisition unit 401, a second acquisition unit 402, a first processing unit 403 and a second processing unit 404.

[0130] Optionally, a first acquisition unit 401 is used to respond to a ticket query request initiated by a user, and to obtain business scenario prompt words, session history and business information through an intelligent agent, wherein the business scenario prompt words are used to characterize the business scenario information corresponding to the intelligent agent, the session history is the historical interaction information between the user and the software system, and the business information is the information related to the aviation business obtained from application services other than the intelligent agent based on the ticket query request and the business scenario information corresponding to the intelligent agent; a second acquisition unit 402 is used to obtain the user's preference information and the status information of each route, wherein the preference information at least includes the user's habit information when handling aviation business and the user's behavioral preference information when selecting aviation flights; a first processing unit 403 is used to generate a basic template based on the business scenario prompt words, and to extract knowledge information from the session history, business information, the user's preference information and the status information of each route, and to fill the knowledge information into the basic template to obtain a target knowledge prompt word; a second processing unit 404 is used to process the ticket query request based on the target knowledge prompt word through a large language model, and to generate at least one recommended ticket information for the user.

[0131] Optionally, the second acquisition unit 402 includes: a first determination subunit, used to determine the user's habit information when handling aviation business and the user's behavioral preference information when selecting aviation flights from the session history and the user's historical itinerary information by means of keyword recognition, so as to obtain the user's preference information; a second determination subunit, used to determine the fare information, flight punctuality rate, flight delay risk information and weather information along the way for each route from the business information; and a third determination subunit, used to use the fare information, flight punctuality rate, flight delay risk information and weather information along the way for each route as the status information of each route.

[0132] Optionally, the first processing unit 403 includes: a first acquisition subunit, used to acquire first-category prompt words, second-category prompt words and third-category prompt words, wherein the first-category prompt words are used to prompt the large language model to use a parsing method when parsing user intentions, the second-category prompt words are used to prompt the large language model to use a reasoning angle when processing target knowledge prompts, and the third-category prompt words are used to constrain the data format filled in the basic template; a first generation subunit, used to generate a basic template based on the first-category prompt words, the second-category prompt words, the third-category prompt words and the business scenario prompt words.

[0133] Optionally, the first processing unit 403 is also used to extract corresponding knowledge information from each of the session history, service information, user preference information and status information of each route using the same aviation business perspective; and / or, to extract corresponding knowledge information from each of the session history, service information, user preference information and status information of each route using different aviation business perspectives.

[0134] Optionally, the air ticket recommendation device based on a large language model further includes: a semantic splicing unit, which is used to perform semantic splicing processing on at least two extracted knowledge information according to the semantic content corresponding to the knowledge information to obtain new knowledge information.

[0135] Optionally, the first processing unit 403 includes: a first processing sub-unit, used to filter the knowledge information extracted from the conversation history, business information, user preference information and the status information of each route according to the business scenario prompt words in the basic template, and obtain the target content whose correlation coefficient with the business scenario prompt words is greater than a preset threshold; a second processing sub-unit, used to fill the target content into the basic template to obtain the target knowledge prompt words.

[0136] Optionally, the first processing unit 403 includes: a third processing sub-unit, which is used to screen the knowledge information extracted from the business information according to the business scenario prompt words in the basic template, and obtain the first content in the knowledge information whose correlation coefficient with the business scenario prompt words is greater than a preset threshold; a fourth processing sub-unit, which is used to integrate the business scenario prompt words and the first content into the second content according to the basic template, and based on the second content, screen the knowledge information extracted from the session history, and obtain the third content in the knowledge information whose correlation coefficient with the second content is greater than a preset threshold; a fifth processing sub-unit, which is used to integrate the second content and the third content into the fourth content according to the basic template, and based on the fourth content, screen the knowledge information extracted from the user's preference information and the status information of each route, and obtain the fifth content in the knowledge information whose correlation coefficient with the fourth content is greater than the preset threshold; a sixth processing sub-unit, which is used to fill the fourth content and the fifth content into the basic template after integrating them, and obtain the target knowledge prompt.

[0137] Optionally, the second processing unit 404 includes: a fourth determination subunit, which is used to perform intent analysis and air ticket information inference on the target knowledge prompt and the air ticket query request according to the first category prompt words and the second category prompt words through a large language model, and determine at least one recommended air ticket information that meets the user's intention based on the predicted user intention and the air ticket information inference result, wherein the expression form of the recommended air ticket information includes at least one of text form, voice form, image form and video form.

[0138] Optionally, the air ticket recommendation device based on the large language model also includes: a first determination unit, which is used to determine, based on the target knowledge prompt, the language, tone and sound style used when conveying the recommended air ticket information to the user through voice when the recommended air ticket information is expressed in voice form; a second determination unit, which is used to determine, based on the target knowledge prompt, based on the large language model, the character form and text expression style used when conveying the recommended air ticket information to the user through text when the recommended air ticket information is expressed in text form; a third determination unit, which is used to determine, based on the target knowledge prompt, based on the large language model, the image elements and the combination of image elements used when conveying the recommended air ticket information to the user through images when the recommended air ticket information is expressed in image form; a fourth determination unit, which is used to determine, based on the target knowledge prompt, based on the large language model, the audio elements and video elements used when conveying the recommended air ticket information to the user through images when the recommended air ticket information is expressed in image form.

[0139] According to another aspect of the present application, a computer-readable storage medium is also provided, wherein the computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned air ticket recommendation method based on the large language model.

[0140] According to another aspect of the present application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the above-mentioned air ticket recommendation method based on a large language model is executed when the program is running.

[0141] According to another aspect of the present application, a computer program product is also provided, comprising computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the above-mentioned air ticket recommendation method based on a large language model are implemented.

[0142] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0143] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0144] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0145] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0146] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0147] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc., which can store program code.

[0148] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for recommending air tickets based on a large language model, characterized in that: include: In response to a flight ticket query request initiated by a user, a business scenario prompt word, a conversation history, and business information are obtained through an agent, wherein the business scenario prompt word is used to characterize the business scenario information corresponding to the agent, the conversation history is historical interaction information between the user and the software system, and the business information is information related to aviation business obtained from application services other than the agent according to the flight ticket query request and the business scenario information corresponding to the agent; Obtaining the user's preference information and status information of each route, wherein the preference information at least includes the user's habit information when handling aviation services and the user's behavior preference information when selecting aviation flights; Generate a basic template according to the business scenario prompt, extract knowledge information from the session history, the business information, the user's preference information and the status information of each route, and fill the knowledge information into the basic template to obtain a target knowledge prompt; The air ticket query request is processed according to the target knowledge prompt through a large language model to generate at least one recommended air ticket information for the user.

2. The air ticket recommendation method based on a large language model according to claim 1, characterized in that: Obtain the user's preference information and the status information of each route, including: By means of keyword recognition, determining the user's habit information when handling aviation services and the user's behavior preference information when selecting aviation flights from the session history and the user's historical itinerary information, and obtaining the user's preference information; Determining the fare information, flight punctuality rate, flight delay risk information and weather information along the route for each route from the business information; The fare information, flight punctuality rate, flight delay risk information and weather information along the route of each route are used as the status information of each route.

3. The air ticket recommendation method based on a large language model according to claim 1, characterized in that: Generate a basic template based on the business scenario prompt words, including: Acquire a first category of prompt words, a second category of prompt words, and a third category of prompt words, wherein the first category of prompt words is used to prompt the large language model to adopt a parsing method when parsing the user's intention, the second category of prompt words is used to prompt the large language model to select a reasoning angle when processing the target knowledge prompt words, and the third category of prompt words is used to constrain the data format filled in the basic template; The basic template is generated according to the first category prompt words, the second category prompt words, the third category prompt words and the business scenario prompt words.

4. The air ticket recommendation method based on a large language model according to claim 1, characterized in that: Extracting knowledge information from the session history, the service information, the user's preference information, and the status information of each route includes: Using the same aviation business perspective, extracting corresponding knowledge information from each of the session history, the business information, the user's preference information, and the state information of each route; and / or, Using different aviation business perspectives, corresponding knowledge information is extracted from each of the session history, the business information, the user's preference information, and the status information of each route.

5. The air ticket recommendation method based on a large language model according to claim 1, characterized in that: After extracting knowledge information from the session history, the service information, the user's preference information, and the status information of each route, the air ticket recommendation method based on the large language model further includes: According to the semantic content corresponding to the knowledge information, semantic concatenation processing is performed on at least two of the extracted knowledge information to obtain new knowledge information.

6. The air ticket recommendation method based on a large language model according to claim 1, characterized in that: Filling the knowledge information into the basic template to obtain the target knowledge prompt includes: Based on the business scenario prompt words in the basic template, information is screened on the knowledge information extracted from the conversation history, the business information, the user's preference information, and the status information of each route to obtain target content with a correlation coefficient with the business scenario prompt words greater than a preset threshold; Fill the target content into the basic template to obtain the target knowledge prompt.

7. The air ticket recommendation method based on a large language model according to claim 1, characterized in that: Filling the knowledge information into the basic template to obtain the target knowledge prompt includes: According to the business scenario prompt words in the basic template, the knowledge information extracted from the business information is screened to obtain the first content in the knowledge information whose correlation coefficient with the business scenario prompt words is greater than a preset threshold; Integrating the business scenario prompt word and the first content into a second content according to the basic template, and filtering the knowledge information extracted from the conversation history according to the second content to obtain a third content in the knowledge information whose correlation coefficient with the second content is greater than the preset threshold; Integrating the second content and the third content into fourth content according to the basic template, and filtering the knowledge information extracted from the user's preference information and the status information of each route according to the fourth content to obtain fifth content in the knowledge information having a correlation coefficient with the fourth content greater than the preset threshold; The fourth content and the fifth content are integrated and filled into the basic template to obtain the target knowledge prompt.

8. The air ticket recommendation method based on a large language model according to claim 3, characterized in that: Processing the air ticket query request according to the target knowledge prompt by the large language model to generate at least one recommended air ticket information for the user includes: The large language model is used to perform intent analysis and air ticket information inference on the target knowledge prompt and the air ticket query request based on the first category of prompt words and the second category of prompt words, and at least one recommended air ticket information that meets the user intent is determined based on the predicted user intent and the air ticket information inference result, wherein the recommended air ticket information is expressed in at least one of text, voice, image and video forms.

9. The air ticket recommendation method based on a large language model according to claim 8, characterized in that: The air ticket recommendation method based on the large language model also includes: When the recommended air ticket information is in the form of voice, the large language model determines the language, tone and voice style to be used when conveying the recommended air ticket information to the user through voice according to the target knowledge prompt; When the recommended air ticket information is in text form, determining, by the large language model and based on the target knowledge prompt, a character form and a text expression style to be used when conveying the recommended air ticket information to the user in text; When the recommended air ticket information is in the form of an image, determining, by the large language model and according to the target knowledge prompt, image elements and a combination of the image elements used when conveying the recommended air ticket information to the user through an image; When the recommended air ticket information is presented in the form of an image, the large language model determines, based on the target knowledge prompt, the audio elements and video elements used when conveying the recommended air ticket information to the user through the image.

10. An air ticket recommendation device based on a large language model, characterized in that: include: A first acquisition unit is used to respond to a ticket query request initiated by a user, and to acquire a business scenario prompt word, a conversation history, and business information through an agent, wherein the business scenario prompt word is used to characterize the business scenario information corresponding to the agent, the conversation history is the historical interaction information between the user and the software system, and the business information is information related to aviation business acquired from application services other than the agent according to the ticket query request and the business scenario information corresponding to the agent; A second acquisition unit is used to acquire the user's preference information and the status information of each route, wherein the preference information at least includes the user's habit information when handling aviation services and the user's behavior preference information when selecting aviation flights; A first processing unit is used to generate a basic template according to the business scenario prompt word, extract knowledge information from the session history, the business information, the user's preference information and the status information of each route, and fill the knowledge information into the basic template to obtain a target knowledge prompt word; The second processing unit is used to process the air ticket query request according to the target knowledge prompt through the large language model to generate at least one recommended air ticket information for the user.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the air ticket recommendation method based on a large language model according to any one of claims 1 to 9.

12. An electronic device, characterized in that: include: A memory storing an executable program; A processor is used to run the program, wherein the program, when running, executes the air ticket recommendation method based on a large language model as described in any one of claims 1 to 9.

13. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by the processor, the steps of the air ticket recommendation method based on a large language model described in any one of claims 1 to 9 are implemented.

Images