Scholarlink

While making flight ticket bookings, customers pose numerous questions about flight policies, ticket fares, baggage policies which they wish to be responded to immediately, and browsing the net or calling the customer support is a gruesome experience and time-consuming. To address these, we have developed a domain specific Large Language Model with Retrieval Augmented Generation with a knowledge base. There are no existing models that can perform the entire booking process using a chatbot. We have developed a chatbot based on OpenAI’s GPT-4o-mini which employs amadeus’ APIs to retrieve and book flight tickets. The system incorporates a vector database for effective information retrieval and MongoDB for storing user and transaction information. Experimental results exhibit considerable improvements in both response time and accuracy compared to conventional customer service processes, with a precision of 0.67 at k \(=\) 3 and a recall of 1.00. This new method provides an end-to-end complete booking experience, from flight search to confirmation, within a conversational interface.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LLM for Airline Assistance Using Retrieval Augmented Generation and MongoDB

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mugesh Raj,
  • Y. Shabanya Kishore,
  • M. Geetha Jenifel

摘要

While making flight ticket bookings, customers pose numerous questions about flight policies, ticket fares, baggage policies which they wish to be responded to immediately, and browsing the net or calling the customer support is a gruesome experience and time-consuming. To address these, we have developed a domain specific Large Language Model with Retrieval Augmented Generation with a knowledge base. There are no existing models that can perform the entire booking process using a chatbot. We have developed a chatbot based on OpenAI’s GPT-4o-mini which employs amadeus’ APIs to retrieve and book flight tickets. The system incorporates a vector database for effective information retrieval and MongoDB for storing user and transaction information. Experimental results exhibit considerable improvements in both response time and accuracy compared to conventional customer service processes, with a precision of 0.67 at k \(=\) 3 and a recall of 1.00. This new method provides an end-to-end complete booking experience, from flight search to confirmation, within a conversational interface.