IntTravel: A Real-World Dataset and Generative Framework for Integrated Multi-Task Travel Recommendation

TL;DR

IntTravel addresses the fragmented landscape of travel recommendation by introducing a large-scale real-world dataset that models four interdependent tasks of a journey: When, Where, How, and Via. It proposes a novel decoder-only multi-task generative framework built from TIP, TSG, and TSF to preserve, filter, and customize information for each task, achieving state-of-the-art results on IntTravel and strong generalization to a non-travel domain. The dataset and framework are validated offline and deployed in Amap, delivering a tangible 1.09% CTR uplift, demonstrating both methodological rigor and practical impact. This work advances holistic journey planning and provides a scalable benchmark for future generative multi-task recommendation research.

Abstract

Next Point of Interest (POI) recommendation is essential for modern mobility and location-based services. To provide a smooth user experience, models must understand several components of a journey holistically: "when to depart", "how to travel", "where to go", and "what needs arise via the route". However, current research is limited by fragmented datasets that focus merely on next POI recommendation ("where to go"), neglecting the departure time, travel mode, and situational requirements along the journey. Furthermore, the limited scale of these datasets impedes accurate evaluation of performance. To bridge this gap, we introduce IntTravel, the first large-scale public dataset for integrated travel recommendation, including 4.1 billion interactions from 163 million users with 7.3 million POIs. Built upon this dataset, we introduce an end-to-end, decoder-only generative framework for multi-task recommendation. It incorporates information preservation, selection, and factorization to balance task collaboration with specialized differentiation, yielding substantial performance gains. The framework's generalizability is highlighted by its state-of-the-art performance across both IntTravel dataset and an additional non-travel benchmark. IntTravel has been successfully deployed on Amap serving hundreds of millions of users, leading to a 1.09% increase in CTR. IntTravel is available at https://github.com/AMAP-ML/IntTravel.

Figures (6)

• Figure 1: The end-to-end pipeline of IntTravel. The workflow processes raw, contextual user logs by tokenizing them into a unified, multi-task sequence format, which includes a shared input sequence and dedicated label sequences for four distinct travel tasks. This structured data is then fed into a decoder-only generative model that jointly predicts all task outcomes, with performance evaluated via a comprehensive multi-dimensional assessment.
• Figure 2: Multi-task framework of IntTravel. IntTravel stacks three core modules to handle multiple tasks. Task-Guided Information Preservation (TIP) ensures task-relevant information is retained. Task-Specific Selective Gating (TSG) filters useful information for each task. Task-Aware Scenario Factorization (TSF) generates task-aware parameters for the output.
• Figure 3: Scaling effects with varying layers. All plotted points are computed after min–max normalization. Task accuracy increases outward while loss decreases outward.
• Figure 4: Proportion of valid data per feature in user profile data of IntTravel. "F1" is short for "Feature 1".
• Figure 5: POI distribution analysis: categorical, administrative, and geographic dimensions. For readability, (b) and (c) only display the top 50 GIDs and ARIDs with the highest number of POIs. Both the geographic and administrative distributions exhibit a long-tail pattern.