How to build recommendation system on Amazon Machine Learning

Question

I wonder if I could build a recommendation system on Amazon Machine Learning. It seems that it's not that easy (according to https://forums.aws.amazon.com/thread.jspa?messageID=617667) and I think if Amazon is the best solution to do that.

Maybe someone has experience with recommendation systems and know how to build it on Amazon ML?

If not, can you suggest me some other SaaS which I can use?

Answer 1

It depends on how many items you need to pick from for your recommendations. You can't use Amazon ML today for recommendations on a huge catalog (such as Amazon.com or other big retailers), but if you have a relatively small catalog (~100 items), you can build a set of binary classifiers for the items.

The quality of a ML model depends on the quality of your data. For example, Amazon.com can make very good recommendations because they have your purchase history for many of your purchases.

A common pattern of recommendation engine is to use collaborative filtering. It can work very well for relatively large number of users and items. This and other similar ML algorithms are implemented in frameworks such as Spark ML (http://spark.apache.org/docs/latest/ml-collaborative-filtering.html). It requires some technical and data science effort to make it work for your use case, but it is usually worth the effort with fairly good results.

You can also try to use some of the following services:

• H2O - http://www.h2o.ai/product/sparkling-water/
• Algorithmia - https://algorithmia.com/algorithms