To the Globe (TTG): Towards Language-Driven Guaranteed Travel Planning

TL;DR

To the Globe (TTG), a real-time demo system that takes natural language requests from users, translates it to symbolic form via a fine-tuned Large Language Model, and produces optimal travel itineraries with Mixed Integer Linear Programming solvers is proposed.

Abstract

Travel planning is a challenging and time-consuming task that aims to find an itinerary which satisfies multiple, interdependent constraints regarding flights, accommodations, attractions, and other travel arrangements. In this paper, we propose To the Globe (TTG), a real-time demo system that takes natural language requests from users, translates it to symbolic form via a fine-tuned Large Language Model, and produces optimal travel itineraries with Mixed Integer Linear Programming solvers. The overall system takes ~5 seconds to reply to the user request with guaranteed itineraries. To train TTG, we develop a synthetic data pipeline that generates user requests, flight and hotel information in symbolic form without human annotations, based on the statistics of real-world datasets, and fine-tune an LLM to translate NL user requests to their symbolic form, which is sent to the symbolic solver to compute optimal itineraries. Our NL-symbolic translation achieves ~91% exact match in a backtranslation metric (i.e., whether the estimated symbolic form of generated natural language matches the groundtruth), and its returned itineraries have a ratio of 0.979 compared to the optimal cost of the ground truth user request. When evaluated by users, TTG achieves consistently high Net Promoter Scores (NPS) of 35-40% on generated itinerary.

Figures (5)

• Figure 1: Front-end interface of TTG. Users send their natural language requests to the demo system (TTG), and TTG replies with itineraries that satisfy user constraints and is optimal with respect to various criteria (e.g., minimal cost).
• Figure 2: Overview of the workflow of TTG. Inference: our system first translates the user travel request in natural language (NL) into the symbolic description of a Mixed Integer Linear programming (MILP) solver using a fine-tuned Large Language Model (LLM), calls the solver to find its optimal solution that satisfies all constraints, and then returns the itinerary in natural language. Training: TTG has three components. A Travel Generator that generates flight/hotel information training data based on real-world data, and symbolic user request $x_i$. An Instruction Translator a pre-trained LLM fine-tuned to translate the NL user request $y_i$ to its symbolic form $\hat{x}_i$, learned by self-consistency between the groundtruth request $x_i$ and the estimated user request $\hat{x}_i$. A Travel solver that solves the estimated symbolic request $\hat{x}_i$ and yields the estimated solution $\hat{s}_i$.
• Figure 3: The breakdown of sources of error in EM accuracy. The three major sources of error are the airline constraints must_not_basic_economy, departure_time, and avoid_red_eye.
• Figure 4: Itinerary options in TTG demo.
• Figure 5: Details of demo. (a) Planned itinerary is shown in tabular view; (b) Flights routes are shown on the map with prices on each travel segment; (c) Hotel infomation, including name, rating and price.