A user trajectory simulation framework for next POI recommendation with uncertain check-ins.

Area of Science:

• * Computational Social Science
• * Data Science
• * Artificial Intelligence

Background:

• * Next point-of-interest (POI) recommendation systems struggle with uncertain check-ins, particularly in collective POIs (CPOIs) like shopping malls.
• * Current methods inadequately represent user behavior within CPOIs, limiting recommendation accuracy.

Purpose of the Study:

• * To introduce TraSim, a novel user trajectory simulation framework designed to enhance next POI recommendation accuracy.
• * To address the challenge of unobserved user behaviors within CPOIs by simulating potential user trajectories.

Main Methods:

• * Utilizes Monte Carlo Tree Search to simulate user trajectories within CPOIs.
• * Incorporates a heuristic reward module integrating IPOI-level signals (temporal traits, transition patterns, user preferences) and CPOI-level semantic constraints.
• * Constructs a multi-candidate trajectory pool to ensure simulation robustness and diversity.

Main Results:

• * TraSim significantly improves recommendation accuracy in uncertain check-in scenarios.
• * Achieved average improvements of 49.1% in Hit Rate (HR) and 44.0% in Mean Reciprocal Rank (MRR) across three real-world datasets.
• * Demonstrates effectiveness and efficiency against nine competitive baselines.

Conclusions:

• * TraSim offers a versatile and scalable solution for next POI recommendation by simulating user behavior within complex venues.
• * The framework is model-agnostic and hyperparameter-free, functioning as a plug-and-play module for existing recommendation systems.