A Visitor‐Centric Recommender System for Florentine Museums: A Foundation for Sustainable Tourism Management

The paper presents the development of a recommender system designed to capture visitors' preferences during museum visits in Florence. The system is trained on data from the entry records of the FirenzeCard, the official museum pass of the city offering tourists access to the Florentine museums. The system utilises an ensemble of deep learning models based on a two-tower architecture, in which separate representations of visitors and museums are learned independently and then combined to generate predictions. It incorporates structured and unstructured data and contextual and visitor features. Results show that the integration of new components into the system architecture significantly enhances the recommender system's ability to accurately predict the top-K museums a visitor is likely to visit next. This aspect makes it one foundational layer of a larger decision-making framework: a reinforcement learning agent to dynamically manage visitor flows, alleviate crowding in overexposed cultural sites and balance urban resources