Similarité globale pour la prédiction de liens dans les réseaux complexes

Abstract

Abstract This thesis addresses the problem of link prediction in complex networks, a critical task for anticipating the emergence of connections between entities. We specifically focus on global similarity methods, which leverage the entire network structure to estimate the likelihood of a link between two nodes. Five methods are studied and compared : Shortest Path, SimRank, Newton’s Gravitational Law Index (NGLI), Katz Index, and Common Neighbor Distance (CND). After presenting the theoretical foundations of graph theory and complex networks, we implemented these methods using Python and applied them to several real-world networks from different domains (biology, transportation, social networks, etc.). The performance of each method was evaluated using standard metrics such as precision, recall, F-measure, and accuracy. The results show that each method has its strengths depending on the network structure, and no single method consistently outperforms the others. This study thus provides valuable insights to guide the choice of link prediction techniques based on specific application contexts.