Apprentissage en temps réel pour la collecte d'information dans les réseaux sociaux

We consider the problem of capturing information on social media under bounded resource. The latter may correspond to real time constraints such as response time limitation, limited computing resources, or social media API restrictions. We formulate this problem as a dynamic source selection problem. We then propose a machine learning methodology for dynamically selecting the most relevant information sources for a given information need. This method is based on an extension of a recently proposed combinatorial bandit algorithm. We provide theoretical guarantees on the behavior of the algorithm. We then evaluate the algorithm on different Twitter datasets for both offline and online settings. MOTS-CLÉS : Apprentissage statistique, réseaux sociaux, bandit manchot