Personalized Parsimonious Language Models for User Modeling in Social Bookmaking Systems

Nawal Ould Amer (1, 2) , Philippe Mulhem (2) , Mathias Géry (3)

1 LIG - Laboratoire d'Informatique de Grenoble
2 MRIM - Modélisation et Recherche d’Information Multimédia [Grenoble]
3 LHC - Laboratoire Hubert Curien [Saint Etienne]

This paper focuses on building accurate profiles of users, based on bookmarking systems. To achieve this goal, we define personalized parsimonious language models that employ three main resources: the tags, the documents tagged by the user and word embeddings that handle general knowledge. Experiments completed on Delicious data show that our proposal outperforms state-of-the-art approaches and non-personalized parsimonious models.

Keywords: User profile; Parsimonious models; Words Embeddings

Domains: Computer Science [cs]; Information Retrieval [cs.IR]

Nawal Ould Amer, Philippe Mulhem, Mathias Géry. Personalized Parsimonious Language Models for User Modeling in Social Bookmaking Systems. European Conference on Information Retrieval, Apr 2017, Aberdeen, United Kingdom. ⟨ujm-01615362⟩