Your point about academia studying different systems than industry rings quite true. Lots of the academic recommender systems are built on datasets and techniq

Your point about academia studying different systems than industry rings quite true. Lots of the academic recommender systems are built on datasets and techniques that simply don't scale to real world data/applications. That said, this pytorch work looks to live more in the realm of application than academia. Building and scaling large neural embeddings is pretty close to industry practice these days and this library at least claims to solve some of the challenges in doing so.