New AI Travel Model Beats Benchmarks, Boosts CTR by 1.09% for 163M Users

Researchers have introduced IntTravel, a massive new dataset and a generative AI framework for integrated travel recommendation. The dataset contains 4.1 billion interactions from 163 million users and 7.3 million points of interest. The decoder-only model holistically plans trips by answering "when to depart," "how to travel," and "where to go." Already deployed on Amap, it achieved state-of-the-art performance and boosted click-through rates by 1.09%.