A fastText-based hybrid recommender - Lateral

Using Facebook Research's new fastText library in supervised mode, I trained a hybrid recommender system, to recommend articles to users, given as training data both the text in the articles and the user/article interaction matrix. The labels attached to a document were both its id, and the ids of all users who viewed it. I've not finished testing it, but early signs are that it learns quite well. This post is a short progress report, perhaps of interest to others building hybrid recommenders. Recently, Facebook Research released the C source code for fastText, a software library for learning word embeddings using a shallow neural network.