Building a recommender engine using Mahout

Question

We have data stored on s3. We want to build a recommender engine using Mahout. Can anybody explain the architecture of building such a system?

Answer 1

Have a look here: https://mahout.apache.org/users/algorithms/recommender-overview.html and here: https://mahout.apache.org/users/algorithms/intro-cooccurrence-spark.html

The best recommender in Mahout is the Spark version of spark-itemsimilarity used with a search engine for fast scalable serving.

The theory behind it allows you to mix many indicators of user taste and to account for context. More concretely this means using many things you know about the user, perhaps multiple actions they have taken, and bias results to fit context like the category of page being viewed.