InsightIQ Blog

Online Analytics in Action - Netflix Competition: Now it’s time to pay up

View more Blog Posts

Jul 1 2009

See a previous post for a summary of the competition - 'Netflix: What were they thinking?

It seems that after 3 years a team has won the prize.  The winning solution was submitted on June 26th, and still must be verified, but all indications are that this solution has broken through a barrier many thought unattainable.  The winning solution predicts customers' movie interests 10%+ better than Netflix's internal system.  That does not sounds like much (10%) but given that Netflix had already squeezed about as much information out of their data as possible, and additional data sources could not be used in the competition, this was a colossal task.   

What's interesting to me about the winning solution is that it is a combination of models.  The top 2 teams decided enough-was-enough and joined forces recently.  They combined their solutions, which individually were close to the 10% mark, and the result was an incremental lift that catapulted them to the finish line. 

This type of analytic solution is called a meta-model, where 2 or more models are combined to enhance performance.  This is always a good practice in customer analytics (marketing analytics, risk analytics, fraud analytics,..) as well as if you have 2 models, which essentially predict the same outcome, combining them will nearly always provide enhanced results.  Even if one model is much better than the other(s), adding models to a solution in effect is adding incremental value not already available in the top model(s).  In fact, every winning solution at the DMA analytic challenge the past 3 years has been the result of a meta-model.

More information on the winning Netflix solution is available in the following link:

http://www.wired.com/epicenter/2009/06/winning-teams-join-to-qualify-for-1-million-netflix-prize/

 

View more Blog Posts

Leave a Comment

  • Your email won't be published on our site.

All Fields Required