Mahout's Flexibility for generating recommendations

Question

I am currently working on a system that will generate product recommendations like those on Amazon: "People who bought this also bought this.."

This is how I plan to do it:

• Process the data using Apache Mahout and generate recommendations(data is stored in MySQL), currently item based only.
• Apply a clustering algorithm to make cluster of users and then apply the association rules subjective to each cluster.

My questions are:

• Will Mahout provide me enough flexibility to tweak the existing algorithms according to my needs. If yes, how?

• Is there a better alternative like R?

Answer 1

As I think we've already discussed in some detail on other threads -- yes you can do things like clustering and collaborative filtering in Mahout. How you do it depends on exactly what you want to do, and it's not entirely clear here. I don't know how you intend to use clustering with recommenders here nor what an association rule is in this context.

But it's open source, so it's the most flexible thing possible: you can make it do whatever you like. So I suppose "yes" is the first answer.

R is not a platform, so I don't think you can view it as somehow an alternative to Mahout.

Answer 2

Yes you can do this. You will need to train several recommenders on each audience segment and have to build your business rules for each segment.

Usually, but not always, this will be a bad idea.

I would also recommend that you go to the Mahout mailing list with these questions. You will find many more Mahout user there.