I'm writing a post for AWS's blog about using Mahout on EMR, since they ship Mahout on their default Hadoop-2-based AMI, and they have a fondness for this project.
They're still shipping Mahout 0.8 so I'll see how hard it is to upgrade the default or at least build another AMI that includes 0.9, as well as how to ask to upgrade future versions. The post uses recommenditembased to build recs for users from the movielens 1M ratings set, then spins up a braindead HTTP API using the Python modules Twisted and Klein along with Redis, and sends requests and gets the responses via curl. Very simple but conceptually traces all the steps of building and shipping a recommender. If you'd like a copy of the pdf version to critique before it goes live let me know.
