Action ML Logo

posted by Pat Ferrel January 15, 2016

A Demo is Worth a Thousand Words


While developing The Universal Recommender we created a demo app. We got real data from reviewers so it makes real recommendations, give it a try. You need to create an account then go to the trainer, which leads you through a bunch of specially chosen videos you can like or dislike. Once you have a few preferences it makes personalized recommendations.

This turned out to be a great experiment because we were able to get lots of user preference indicators. You might not guess it but we found user dislikes helped predict likes.

This demo uses ActionML's Universal Recommender for multi-action recommendations and Clustering to pick the most differentiating videos for the trainer. We illustrate several UR features here like the use of video properties to narrow down recommendations in a hybrid of Collaborative Filtering and Content-based recommendations.

Caveat Emptor: Sorry for dupes in the Videos list, RT could use a little data scrubbing. Don’t expect much from search it is only minimally hooked up for how.