I don't get it. How is it different from "Wide & Deep Learning for Recommender Systems (2016)" by Google?<p>The whole project, up til now, contains less than 2,500 lines of Python, and 271 lines of Markdown as doc, 159 lines of Shell script, and a Kaggle benchmark on Criteo CTR which dated back to 2014. The project seems to be published by FB Research, while I do not see much "research" value in this project or the companion blog.