Word2Vec with Spark ML on the 20 Newsgroups dataset