Abstract
News recommender systems are aimed to personalize users experiences and helpthem to discover relevant articles from a large and dynamic search space.Therefore, news domain is a challenging scenario for recommendations, due toits sparse user profiling, fast growing number of items, accelerated item'svalue decay, and users preferences dynamic shift. Some promising results havebeen recently achieved by the usage of Deep Learning techniques on RecommenderSystems, specially for item's feature extraction and for session-basedrecommendations with Recurrent Neural Networks. In this paper, its presented aDeep Learning architecture for Session-Based recommendations of News articles.This architecture is composed of two modules, the first responsible to learnnews articles representations, based on their text and metadata, and the secondmodule aimed to provide session-based recommendations using Recurrent NeuralNetworks. The recommendation task addressed in this work is next-itemprediction for user sessions: "what is the next most likely article a usermight read in a session?" User session context is leveraged by the architectureto provide additional information in such extreme cold-start scenario of newsrecommendation. Users' behavior and item features are both merged in an hybridrecommendation approach. A temporal offline evaluation method is also proposedas a complementary contribution, for a more realistic evaluation of such task,considering dynamic factors that affect global readership interests likepopularity, recency, and seasonality.