Alternatives (or ways to speed up) Acts_As_Recommendable plugin for Ruby on Rails

Question

Hi all-

I am currently using the Acts_as_recommendable plugin available here. It is using the pearson correlation coefficient to find recommendations, which is pretty much exactly what I want. The problem however is scale. With more than 2000 or so items, the plugin slows considerably (with 5000 items, I see load times of about a minute for a recommended_items call).

So my question: are there any ruby/rails alternatives that are viable? I have tried working with mahout, but it is fairly hard to maintain and having both an amazon elastic compute instance and a heroku instance doubles that pain.

Thanks!

Answer 1

Have you been able to use a Cached Dataset for AAR?

In production, you could User.aar_options[:use_dataset] = true

And then in a cron job every so often: rake task recommendations:build

I'm unsure on the ability to use heroku for this though -- try it out and post back :)

Answer 2

(I am an author of the Mahout recommenders.) Working with jobs on Hadoop is difficult indeed. If that's your main issue with Mahout, then might I point out that it has significant non-distributed, non-Hadoop-based recommender support. My rule of thumb is that it outgrows a single box when you hit about 100M ratings, which is way larger than your current size.