Passé, présent, futurs : induction de carrières professionnelles à partir de CV

Extracting, structuring and exploiting information from freeform text is a difficult task. Learning embeddings with chosen properties and going beyond simple syntax encoding contributed to significant improvements in semantic analysis. Recently, the focus has shifted from word and document embeddings to reasoning in order to infer or predict new knowledge. In this paper, we focus on job & educational background embeddings that are learned from a large CV corpus. We aim at modeling users careers and forecasting their choices. Inspired by recent work in machine translation, we design a Recurrent Neural Network architecture to normalize job & qualification titles. Once this semantic step achieved, we build another RNN to predict position chaining in CV. MOTS-CLÉS : Apprentissage de représentations, Réseaux de neurones, Recommandation.