Tourism AI assistant system and method based on multi-agent cooperation

Abstract

The invention discloses a tourism AI assistant system and method based on multi-agent cooperation, and the system comprises a front-end user terminal which is configured to provide a user interface, the user interface comprises a chat interaction area and a map display area, the chat interaction area is used for receiving the natural language input of a user and displaying the dialogue information returned by a rear end, and the map display area is used for displaying the dialogue information; the map display area is used for dynamically visualizing geographical location information; the back-end server is in communication connection with the front-end user terminal, a multi-agent system is deployed in the back-end server, and the multi-agent system comprises a main agent and a plurality of hierarchical sub-agents; and the main agent is configured to receive and analyze natural language input sent by the front-end user terminal to identify the travel intention of the user, and dispatch the corresponding sub-agents according to the identified travel intention of the user. According to the system and the method, one-stop, intelligent and dynamic tourism planning service is realized, and the user experience and the planning efficiency are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence technology and tourism service integration, in particular to a tourism AI assistant system and method based on multi-agent collaboration. BACKGROUND

[0002] With the digital development of the tourism industry, users' demand for intelligent and convenient tourism planning tools is increasing. Currently, the existing technology has the following significant defects: Information and services are fragmented: users need to switch between multiple applications or websites to query destinations, book transportation and accommodation, information cannot be effectively integrated, and operations are cumbersome.

[0003] Interaction mode is rigid: relying on structured forms or menu selection, lacking natural language dialogue capability, difficult to support continuous, context-related complex planning tasks.

[0004] Insufficient personalization and intelligence: recommendations based on historical browsing or general tags, unable to understand users' immediate and detailed demand descriptions, low recommendation accuracy.

[0005] Planning process is linear: forcing users to follow a fixed process (such as first determining the destination and then determining the transportation), unable to support dynamic adjustment, and disconnected from actual decision-making logic.

[0006] Therefore, there is an urgent need for a tourism assistant system that can integrate services, support natural interaction, and dynamic planning to solve the above problems. SUMMARY

[0007] To solve the technical problems in the background art, the present application proposes a tourism AI assistant system and method based on multi-agent collaboration.

[0008] The tourism AI assistant system based on multi-agent collaboration proposed by the present application comprises: A front-end user terminal configured to provide a user interface, the user interface including a chat interaction area and a map display area, the chat interaction area for receiving user natural language input and displaying dialog information returned by the back-end, and the map display area for dynamically visualizing geographic location information; A back-end server in communication connection with the front-end user terminal, the back-end server internally deploying a multi-agent system, the multi-agent system including a master agent and a plurality of hierarchical sub-agents; The master agent is configured to receive and analyze natural language input sent by the front-end user terminal to identify user travel intentions, and to dispatch corresponding sub-agents according to the identified user travel intentions; The sub-agent is configured to perform specific tasks according to the dispatch of the master agent, including obtaining tourism-related data and returning to the master agent; The main agent is further configured to integrate the data returned by the sub-agents, generate a natural language reply and transmit to the front-end user terminal.

[0009] Preferably, the sub-agents include inspiration-inspired agents and travel planning agents; the inspiration-inspired agents include a place recommendation sub-agent and a point of interest recommendation sub-agent, and the travel planning agents include a flight search sub-agent, a hotel search sub-agent and a travel itinerary generation sub-agent.

[0010] Preferably, the place recommendation sub-agent and the point of interest recommendation sub-agent obtain place information by calling external map APIs, the flight search sub-agent and the hotel search sub-agent automatically crawl real-time data from online platforms through network automation tools, and the travel itinerary generation sub-agent obtains historical dialogue information through a memory tool and generates a structured travel itinerary.

[0011] Preferably, the main agent parses the natural language input based on a large language model, and can call the sub-agents multiple rounds to obtain supplementary information when the information is insufficient.

[0012] Preferably, the main agent sends the natural language reply to the front-end user terminal in a streaming manner through a server push event technology, the front-end user terminal receives and renders to a chat interaction area in real time after receiving, and parses geographic location coordinates to mark on a map display area.

[0013] The present application proposes a kind of tourism AI assistant method based on multi-agent cooperation, comprising the following steps: The front-end user terminal receives user natural language input and sends to back-end server; The main agent of back-end server parses the natural language input to identify user travel intention; The main agent dispatches corresponding sub-agent to execute task according to the identified user travel intention; The sub-agent executes task and returns the tourism related data obtained to the main agent; The main agent integrates the tourism related data, generates a natural language reply and transmits to the front-end user terminal; The front-end user terminal displays the natural language reply, and visualizes related geographic location information in map display area.

[0014] Preferably, the sub-agent executing task includes: The place recommendation sub-agent recommends tourist attractions according to season and user label and calls external map API to obtain details; The point of interest recommendation sub-agent recommends scenic spots, restaurants or activities in destination; The flight search sub-agent and the hotel search sub-agent automatically capture real-time flight and hotel data through network automation tools. The itinerary generation sub-agent integrates historical information to generate a structured itinerary.

[0015] Preferably, the main agent utilizes a large language model to understand complex instructions related to context when parsing natural language input, and triggers multiple rounds of sub-agent calls when data does not meet requirements.

[0016] Preferably, the natural language reply generated by the main agent is sent to the front-end user terminal in Markdown format through streaming, and the front-end user terminal updates the chat interaction area and map display area synchronously after parsing.

[0017] In the present application, the proposed tourism AI assistant system and method based on multi-agent collaboration include a front-end user terminal and a back-end server. The front-end user terminal provides a user interface containing a chat interaction area and a map display area, enabling natural language interaction and geographic location visualization. The back-end server deploys a processing system based on a multi-agent architecture, which parses user intent through a main agent and dispatches hierarchical sub-agents. The sub-agents obtain real-time data by calling external APIs and network automation tools, and return the data to the front-end through streaming after integration by the main agent. The present application solves the problems of fragmented tourism information, rigid interaction, insufficient personalization, and linear planning in the prior art, and provides one-stop, intelligent, and dynamic tourism planning services, significantly improving user experience and planning efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0018] Fig. 1 The present application proposes a whole framework diagram of a tourism AI assistant system based on multi-agent collaboration; Fig. 2 The present application proposes a working process diagram of a tourism AI assistant method based on multi-agent collaboration; Fig. 3 The present application proposes an implementation process diagram of a tourism AI assistant method based on multi-agent collaboration. DETAILED DESCRIPTION

[0019] REFERENCE Figs. 1-3 The present application proposes a tourism AI assistant system based on multi-agent collaboration, which includes: The front-end user terminal is configured to provide a user interface containing a chat interaction area and a map display area. The chat interaction area is used to receive user natural language input and display dialog information returned by the back-end. The map display area is used to dynamically visualize geographic location information.

[0020] Specifically, natural language dialogue, maps, information queries, and itinerary planning are seamlessly integrated into a single interface, avoiding the hassle of switching between multiple applications and providing a smooth and coherent one-stop service. After the front-end user terminal receives the Markdown-formatted streaming data, it renders it in real-time to the chat interaction area. At the same time, it parses the geographic location coordinates (latitude and longitude) contained in the reply and positions and highlights them on the map display area.

[0021] The backend server is in communication connection with the front-end user terminal, and the backend server internally deploys a multi-agent system, which includes a master agent and multiple hierarchical sub-agents.

[0022] Specifically, through the multi-agent collaboration architecture, the system can deeply understand complex, context-related natural language instructions and autonomously call corresponding tools to complete tasks, realizing a mode shift from "people looking for tools" to "AI-centered."

[0023] The master agent is configured to receive and parse natural language input sent by the front-end user terminal to identify user travel intentions, and to dispatch corresponding sub-agents according to the identified user travel intentions; The sub-agents are configured to perform specific tasks according to the dispatch of the master agent, including obtaining travel-related data and returning it to the master agent; The master agent is also configured to integrate the data returned by the sub-agents, generate a natural language reply, and transmit it to the front-end user terminal.

[0024] In this embodiment, the sub-agents include inspiration-inspired agents and travel planning agents; the inspiration-inspired agents include a place recommendation sub-agent and a point of interest recommendation sub-agent, and the travel planning agents include a flight search sub-agent, a hotel search sub-agent, and a trip generation sub-agent.

[0025] In this embodiment, the place recommendation sub-agent and the point of interest recommendation sub-agent obtain place information by calling external map APIs, the flight search sub-agent and the hotel search sub-agent automatically scrape real-time data from online platforms through network automation tools, and the trip generation sub-agent obtains historical dialogue information through a memory tool and generates a structured itinerary.

[0026] Specifically, by calling real-time APIs and using network automation scraping techniques, the timeliness and accuracy of information such as places, flights, and hotels are ensured, which is superior to traditional solutions that rely on static data.

[0027] In this embodiment, the master agent parses natural language input based on a large language model, and can call sub-agents multiple times to obtain supplementary information when information is insufficient.

[0028] In the embodiment, the main agent sends the natural language reply to the front-end user terminal in a streaming manner through a server push event technology, the front-end user terminal receives and renders to a chat interaction area in real time, and a geographic position coordinate is parsed and marked on a map display area.

[0029] Specifically, the system has a memory function, can dynamically adjust the planning content according to the continuous dialogue, and can actively recommend the next operation in combination with the context, so that the whole planning process is closer to the natural decision-making thinking of human beings.

[0030] In the embodiment, the location recommendation sub-agent is used to recommend suitable tourist cities or regions according to conditions such as seasons and user tags, and to obtain specific information of the location by calling external APIs such as Gaode Map. The point of interest recommendation sub-agent is used to recommend local scenic spots, restaurants or activities after the destination is determined, and also relies on external map APIs to obtain detailed information (such as introduction, pictures, score). The flight search sub-agent is used to be activated when the user shows the intention to query the flight, and to simulate user operations by using network automation tools such as Playwright to real-time capture flight schedules, prices and other information from one or more specified online ticket platforms. The hotel search sub-agent is used to filter and capture the matching hotel list from online booking platforms according to the user's set destination, date, budget and other conditions, by using network automation tools. The itinerary generation sub-agent is used to integrate the information of the user-confirmed destination, flight, hotel and the like in the foregoing steps to generate a structured itinerary in JSON format. The agent has a memory function, and can remember the key choices in the multi-round dialogue through the memory tool of the ADK framework, to provide complete context for the final itinerary planning.

[0031] In the embodiment, the main agent is also configured to make judgments and decisions after receiving the execution results of the sub-agents. If the current information is insufficient to answer the user's question, it can call the same or another sub-agent again to obtain supplementary information (i.e. multi-round calling). When the decision terminates the calling, the main agent will integrate all the collected information, organize a coherent and complete natural language reply, and send it to the front-end user terminal in a streaming manner through the server push event (SSE) technology.

[0032] Referring Figs. 1-3 The application provides a tourism AI assistant method based on multi-agent cooperation, which comprises the following steps: The front-end user terminal receives the natural language input of the user and sends it to the back-end server; The main agent of the back-end server parses the natural language input to identify the tourism intention of the user; The main agent dispatches corresponding sub-agents to execute tasks according to the identified tourism intention of the user; The sub-agent performs a task and returns the obtained travel-related data to the main agent; The main agent integrates the travel-related data, generates a natural language reply, and transmits it to the front-end user terminal; The front-end user terminal displays the natural language reply and visualizes the relevant geographic location information in the map display area.

[0033] In this embodiment, the task performed by the sub-agent includes: The place recommendation sub-agent recommends tourist attractions according to the season and user tags and calls external map API to obtain details; The point of interest recommendation sub-agent recommends scenic spots, restaurants, or activities in the destination; The flight search sub-agent and the hotel search sub-agent automatically extract real-time flight and hotel data through network automation tools; The itinerary generation sub-agent integrates historical information to generate a structured itinerary.

[0034] In this embodiment, the main agent uses a large language model to understand complex instructions related to context when parsing natural language input, and triggers multiple rounds of sub-agent calls when the data does not meet the requirements.

[0035] In this embodiment, the natural language reply generated by the main agent is sent to the front-end user terminal in Markdown format through streaming, and the front-end user terminal updates the chat interaction area and the map display area synchronously after parsing. Embodiment

[0036] Step 1: The system receives the user's natural language input "I want to go to Xinjiang next month, what do you recommend?"

[0037] Step 2: The main agent parses the natural language input and identifies that the user's travel intention is "destination recommendation", with keywords including "next month" and "Xinjiang".

[0038] Step 3: The main agent decides to call the inspiration-inspired agent under it, and further activates the place recommendation sub-agent (place_agent).

[0039] Step 4: The place recommendation sub-agent performs a task, queries the popular tourist attractions related to "Xinjiang" and "next month" (which can be associated with seasonal characteristics) through the LLM large model, such as "Kashgar" and "Heavenly Lake", then calls external map service API to obtain the introduction, pictures, and ratings of these places, and returns these structured data to the main agent.

[0040] Step 5: The main agent receives the returned result. It uses the LLM to organize the structured data into a piece of natural and fluent recommendation language, and outputs it back to the front end through the SSE stream. The front end displays it in the chat interaction area, and at the same time marks the positions of Kashi and Tianchi on the map. The main agent also generates a suggestive question based on the context, such as "Which place are you more interested in? Or do you want to know the flight information to these places?", assuming that the user then asks "Check the flights from Hangzhou to Urumqi in the middle of next month".

[0041] Step 6: The main agent identifies the new intent "flight search".

[0042] Step 7: The main agent decides to call the travel planning agent, and activates the flight search sub-agent (flight_search_agent) under it.

[0043] Step 8: The flight search sub-agent starts the network automation tool (such as Playwright) through the interface. Through this tool, it automatically accesses one or more online ticketing websites, automatically inputs "Hangzhou" to "Urumqi" and the corresponding date, grabs the flight list including flight number, departure and arrival time, airline and ticket price, and returns the grabbing result to the main agent.

[0044] Step 9: The main agent formats the flight information and returns it to the front end for display.

[0045] Step 10: After receiving the data returned by the back end, the front end parses the geographic location information in it, then renders the markdown content of the reply to the chat box on the left side, and marks the places mentioned in the reply content on the right side map.

[0046] The process of hotel search and itinerary generation is similar. When the user selects the flight and hotel, the itinerary generation sub-agent (itinerary_agent) is activated. It can access the information determined in the previous conversation (such as destination: Urumqi, selected flight, selected hotel) through the memory tool provided by the ADK framework, integrate these fragmented information, and generate a complete itinerary, which is displayed to the user in a beautiful way.

[0047] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A multi-agent collaboration-based tourism AI assistant system, characterized in that, The application relates to a system and method for providing natural language-based travel information, comprising: a front-end user terminal configured to provide a user interface, which includes a chat interaction area for receiving user natural language input and displaying conversational information returned by a back-end server, and a map display area for dynamically visualizing geographic location information; a back-end server in communication connection with the front-end user terminal, which internally deploys a multi-agent system including a master agent and multiple hierarchical sub-agents; the master agent is configured to receive and analyze the natural language input sent by the front-end user terminal to identify the user's travel intention, and dispatch corresponding sub-agents according to the identified user's travel intention; the sub-agents are configured to perform specific tasks according to the dispatch of the master agent, including obtaining travel-related data and returning to the master agent; the master agent is also configured to integrate the data returned by the sub-agents, generate a natural language reply and transmit it to the front-end user terminal. 2.The multi-agent collaboration-based tourism AI assistant system according to claim 1, characterized in that, The sub-agents include inspiration-inspired agents and travel planning agents; the inspiration-inspired agents include a place recommendation sub-agent and a point of interest recommendation sub-agent, and the travel planning agents include a flight search sub-agent, a hotel search sub-agent and a travel itinerary generation sub-agent. 3.The multi-agent collaboration-based tourism AI assistant system according to claim 2, characterized in that, The place recommendation sub-agent and the point of interest recommendation sub-agent obtain place information by calling external map APIs, the flight search sub-agent and the hotel search sub-agent automatically capture real-time data from online platforms through network automation tools, and the travel itinerary generation sub-agent obtains historical dialogue information through a memory tool and generates a structured travel itinerary. 4.The multi-agent collaboration-based tourism Al assistant system according to claim 1, characterized in that, The master agent analyzes the natural language input based on a large language model, and can call sub-agents multiple times to obtain supplementary information when the information is insufficient. 5.The multi-agent collaboration based tourism Al assistant system according to claim 1, wherein, The master agent sends the natural language reply to the front-end user terminal in a streaming manner through server push event technology, the front-end user terminal receives and renders it to the chat interaction area in real time, and analyzes the geographic location coordinates to mark them on the map display area. 6.A method for a tourism AI assistant based on multi-agent cooperation, characterized in that, The application also provides a method for providing natural language-based travel information, comprising the following steps: the front-end user terminal receives user natural language input and sends it to the back-end server; the master agent of the back-end server analyzes the natural language input to identify the user's travel intention; the master agent dispatches corresponding sub-agents to perform tasks according to the identified user's travel intention; the sub-agents perform tasks and return the obtained travel-related data to the master agent; the master agent integrates the travel-related data, generates a natural language reply and transmits it to the front-end user terminal; the front-end user terminal displays the natural language reply and visualizes the related geographic location information in the map display area. 7.The multi-agent collaboration-based tourism AI assistant method according to claim 6, characterized in that, The task performed by the sub-agents includes: the place recommendation sub-agent recommends travel places according to seasons and user tags and calls external map APIs to obtain details; the point of interest recommendation sub-agent recommends scenic spots, restaurants or activities in the destination; the flight search sub-agent and the hotel search sub-agent capture real-time flight and hotel data through network automation tools; the travel itinerary generation sub-agent integrates historical information to generate a structured travel itinerary. 8.The multi-agent collaboration-based tourism AI assistant method of claim 6, wherein, The main intelligent agent utilizes a large language model to understand complex instructions related to context when analyzing natural language input, and triggers multiple rounds of sub-agent calls when data does not meet requirements. 9.The multi-agent collaboration-based tourism AI assistant method of claim 6, wherein, The natural language reply generated by the main intelligent agent is sent to the front-end user terminal in Markdown format through streaming, and the front-end user terminal updates the chat interaction area and map display area synchronously after parsing.

Images