TempoBiGen: A Curated Generative Model for Healthcare Mobility Logs with Visit Duration

Abstract

Healthcare facilities, which serve vulnerable populations such as patients and the elderly, are also hotspots where pathogens drive nosocomial infections. Accurately modeling pathogen transmission within these settings is essential for understanding their dynamics and enhancing preparedness and intervention strategies. A key barrier to achieving high-fidelity models of pathogen transmission within healthcare facilities is the scarcity of fine-grained, high-quality mobility logs that capture real-world interactions. Data synthesis offers a promising solution by generating realistic mobility datasets. Existing methods can generate synthetic mobility logs while preserving the temporal evolution of the original data’s structural properties. However, these approaches are limited in two key ways: (1) they typically overlook the bipartite structure inherent in mobility logs (e.g., interactions between healthcare workers and rooms), and (2) they fail to account for the duration of interactions, a critical factor in transmission dynamics. Building on top of existing work, we introduce TempoBiGen , a curated generative model designed to address these shortcomings. TempoBiGen explicitly models bipartite temporal networks and incorporates visit duration in a post-processing step, producing high-fidelity, ready-to-use synthetic mobility logs. We evaluated TempoBiGen using real-world mobility logs gathered from healthcare facilities, assessing its performance in preserving snapshot-based graph properties (e.g., degree distribution and connected components size) and replicating temporal dynamics through disease spread simulations. Our results demonstrate that the proposed approach leads to a robust and effective tool for generating synthetic mobility data, offering additional resources to enhance modeling and analyzing hospital mobility patterns.