Modèle de graphe hiérarchique pour la représentation et l’analyse de la mobilité : application au trafic maritime
Soutenue par Maryam Maslek Elayam, à l’Université de Bretagne Occidentale le 28/11/2024. Encadrée par Christophe Claramunt, Cyril Ray. Grand plaisir d’avoir pu échangé avec l’équipe de l’école navale sur un sujet passionnant . Lien [Rapporteur]
Mobility representation and analysis offer a structured approach to better understand move- ment behaviors and the spatial, temporal, and semantic dynamics within complex networks. Modeling mobility data and trajectories using graphs allows for extracting topological, structural, and dynamic analyses of the transport network, relying on the fundamentals and algorithms proposed by graph theory. In this thesis, we propose a hierarchical graph model with different levels of abstraction to represent spatio-temporal and semantic trajectories. Our model provides a multi-scale view that enables analyses from fine scale to regional and global scales of patterns, movement flows, and economic exchanges. We present a prototype of a hierarchical graph database using three levels of abstraction, applied to maritime mobility data. Exploring the potential of our model gives us the opportunity to develop queries and graph operators tailored to the formal conception of our hierarchical model. In this context, we develop two metrics: hierarchical accessibility to characterize the topology of ports within the maritime network, and hierarchical similarity to examine structural changes over time. Futhermore, we focus on analyzing emerging structures within the hierarchical graph using deep learning methods, specifically graph convolutional networks, to address the problem of detecting ports communities in the hierarchical network.