New algorithm - better personalization, more serendipity

The new recommendation algorithm we've been working on is now live for all outbrain users[1]. With the new algorithm we tried to touch one main point - better personalization. Our previous algorithm would find similarities between outbrainers based on a small number of similar positive votes. Once two users were associated by votes, it was difficult for those two users to drift apart. It's like two random people in a theater who happened to like the movie, finding out as they're exiting the theater that they are now married on that merit...

The most noticeable effect of this 'happened-to-like-the-same-movie' syndrome, was that recommendations became very monotonous and limited to those few blogs that your newlywed spouse happened to like... ;-)

The new algorithm is much more dynamic and sensitive to people's similarities. This means that every time you vote, whether high or low, outbrain will adjust itself to find those people with the most similar interests to you.

Beyond the much better personalization, you should notice much more serendipity in the recommended items. This is something that should further improve with each new user we add into the system.

We'd love to get your feedback on the new algorithm. The best possible feedback is simply by voting on your recommended items every day. We look closely at those votes and tweak the algorithm settings constantly (+of course, the more you vote, the smarter your personal outbrain will get!).

Any other feedback? Shoot us an email to:
info (at) outbrain (dot) com

[1] If you haven't subscribed to your personal recommendation feed yet, do so by pointing your RSS reader to the following location (don't forget to change the USERNAME part with your outbrain user name):

http://dsf.outbrain.com/dsf/USERNAME.xml