Détection de nouveauté dans des flux de données textuelles
Lundi 17/02/2020, 11:00, Salle K71
Detecting novel topics and/or documents in textual data streams is a challenge for several companies. Indeed, EDF (Electricité de France), the French national electricity producer, wants to be able to optimize its marketing responses by being aware at the earliest of the emergent topics discussed about them in corpora of customer complaints or email. We will define more precisely the concept of novelty and we will see that a major challenge with this kind of task lies in the evaluation methods. Finally, we will discussed approach based on probabilistic models like Latent Dirichlet Allocation, forecasting models and change detection methods.