Increasing Relevancy of Recommendations with Human Pyramid – Part 2

In the previous post, I proposed a six layer human pyramid that could overlay data-driven relevancy and discussed the bottom 3 levels as they relate to a ‘Recent Releases’ dvd movie website.  In this post, I’ll describe the top 3 layers, which are potentially the most accurate filters, but can be difficult to accurately achieve.  For reference, here is the pyramid again:

Recommendation Pyramid - Personalized, Social, Community

The bottom 3 layers focus on other people: what critics like, what the community likes, what my friends like.  The top 3 layers shift away from this and move the focus to me.  They are personalized to my tastes, which adds a layer of friction to the service, as I must first provide information about myself to the site before it can use these layers as filters to better serve up relevant results.  Each layer going up needs more data about me than the last to be successful.

People I Trust – This differs from friends for two reasons.  The first is that there may be people I trust that aren’t my friends, whether they be critics that I agree with or other community members that have similar opinions to me.  The second is that my friends don’t always have the same tastes as me.  My best friend may be a classic movie buff while I like lighthearted comedies and popcorn blockbusters.  If the movie site allows me to identify users and critics that I agree with and trust, then it can more accurately pare the 90 movies down to the handful that I’d enjoy.

People Like Me – The site can track my movie ratings over time and compare those to other users who rated a number of the same movies.  The users whose scores have a high correlation to my own can be said to be similar to me.  It is likely that if someone that has similiar tastes to mine likes a specific movie that I will also like that movie.  The more data the site has on me and the similar users, the more accurate this filter can be.  Flixster also leverages this functionality, even making a popular facebook game out of it, matching people with their friends that are most compatible with them movie-wise.

Me – The ultimate relevancy filter would be able to know everything about me and use that to pinpoint the exact movies that I would like to see.  This would include knowing the movies on the list that I’ve already seen and that I only have 2 hours to watch a movie, so it can exclude the movies I’ve seen and movies longer than 120 minutes from the results.  The trick is how to get this data from users.  Netflix will recommend movies to you based on the movies you’ve rated in their system.  The more movies that you rate, the more accurate the recommendations will be, so this service gets more valuable the more you use it.

One Response to “Increasing Relevancy of Recommendations with Human Pyramid – Part 2”

  1. […] This post was mentioned on Twitter by Nicholas P. Nicholas P said: Increasing Relevancy of Recommendations with Human Pyramid – Part 2 […]

Leave a Reply