Algorithmes de bandit pour les systèmes de recommandation : le cas de multiples recommandations simultanées

The multiple-plays recommender systems (RS) make a refer to RS which recommend several items to the users. Most of RS are based on learning models in order to choose items to be recommended. Among learning models, the bandit algorithms offer the advantage to learn and exploite the already learnt elements at the same time. The current approaches require running as many instance of single-play bandit algorithms as there are items to recommend. We propose to manage all recommendations simultaneously, by a single instance of one single-play bandit algorithm. We show on two benchmark datasets (Movielens and Jester) that our method, MPB (Multiple Plays Bandit), allows to obtain a learning rate until thirteen times faster while obtaining equivalent click-through rates. We also show that the choice of the bandit algorithm used impacts this improvement. MOTS-CLÉS : Recherche d’Information, Systèmes de Recommandation, Problème de Bandits.