In this work we focus on the use of distributed representations of words and documents using the continuous Skip-gram model. We compare this model with three recent approaches: Information Gain Word-Patterns, TF-IDF graphs ...