LinkedIn Twitter
CMO at SecureAuth Corporation. Previously VP Marketing at NowSecure, Knurld, Ping Identity, VP Business Develoment at Get Satisfaction, CEO at Teqlo and General Partner at SAP Ventures.

2 responses to “Measuring Popularity in Communities”

  1. Ernie

    Jeff,

    Those are not the actual ratings on the Netflix website, the ones in red are the ratings they predict for you and your tastes. Which by your evaluation of the titles they provided to you, shows their matching algorithms to your tastes to be some what accurate. Their matching algorithms are based on your previous ratings to similar content movies. If you have not rated movies in the past or share your account with a spouse or child and they do rate movies, then the rsulting recommendation for you in particular may be skewed.

    “I think where I come down on this is that Netflix should be ranking movie popularity according to customer reviews and applying a weighting or logarithmic function that normalizes the curve to allow new movies to float up the ranking ladder to avoid “the more you have the more you get” phenomena that voting systems often fall into.”

    I’m sure they do have a weighted algorithm that considers other reviewers that rated movies similar to what you rated a movie. That being said their design is probably dependent upon only one person on each account giving the reviews. There is also things to take into consideration that may change how someone rates similar items. I do not think “American Pie” is a very good movie now but at the time when I first saw it at 19 I thought it was hilarious. That being said when the algorithm is developed that accounts for social norm changes in society, changing personal preferences, life altering events, and mental development. You may have a Billion dollar algorithm all companies with a marketing department would kill to get a hold of.

    Thanks,

    Ernie

  2. Jeff Nolan

    Thanks Ernie, good information.

    The problem here is that popularity is a way of filtering and should never display recommendations that are “popular but poorly rated”. This is where the semantics of Netflix’s approach collides with how average users prioritize the content being provided to them.