My employer owns different regional newspapers, local tv stations and special-interest magazines.

In a way to increase page views and usage time, we would like to make the recommendations to our users ("you may also like...") as relevant as possible.

For that we're looking for a recommender system that works throughout our pages. Our requirements:

  • Collect data as implicit as possible: We don't want users to star-rate content, but would rather collect what and/or how long they read/watch our content.
  • The context of our content is always an article
  • A minimal article has a title, a lead and a body (which can be text, a video or an image gallery)
  • Most of our users are anonymous resp. can't be matched to a UUID
  • We love Ruby, so a ruby-based recommender system would be appreciated (but is not a prerequisite)
  • Should run on Linux.

What kind of collaborative filtering, content-based filtering or hybrid recommender system would you use in our case and why?

We're currently looking at Easyrec and Recommendable but are very much open to new suggestions.


1 Answer 1


I am not an expert on FOSS recommender systems - I have used more than one to do Cross-Selling, but all of them built from scratch.

Recommendable does not appear to be applicable to your case. Recommendable sends the recommendation after the user has rated the current item. Since you are scoring the items in a way that doesn't involve user interaction, you have score it based on clicks and time - if the user spends more than a minute on the article before navigating away, the article gets a like. You can use javacript to send the like after a minute has passed, and then receive back a list of recommendations. But this way recommendable never gets a "dislike" rating (and making it aware that the user navigated away would interfere with the user experience).

The FOSS recommender that attracted me the most is prediction.io, a JVM based system. It needs to keep track of UUIDs, though.

  • 1
    PS.: I will be rolling out my own proprietary general-purpose recommendation engine in a a few months. While our focus right now is location and sentiment aware recommendation, I will try to address this question on it. Commented Mar 27, 2014 at 13:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.