Whitepaper on TiVo's filtering
Every wonder how the collaborative filtering system works on a TiVo? Wijnand van Stam from TiVo co-authored a detailed paper describing how it all works (word doc) (google's html version of the word doc).
The abstract:
We describe the TiVo television show collaborative recommendation system which has been fielded in over one million TiVo clients for four years. Over this install base, TiVo currently has approximately 100 million ratings by users over approximately 30,000 distinct TV shows and movies. TiVo uses an item-item (show to show) form of collaborative filtering which obviates the need to keep any persistent memory of each user�s viewing preferences at the TiVo server. Taking advantage of TiVo�s client-server architecture has produced a novel collaborative filtering system in which the server does a minimum of work and most work is delegated to the numerous clients. Nevertheless, the server-side processing is also highly scalable and parallelizable. Although we have not performed formal empirical evaluations of its accuracy, internal studies have shown its recommendations to be useful even for multiple user households. TiVo�s architecture also allows for throttling of the server so if more server-side resources become available, more correlations can be computed on the server allowing TiVo to make recommendations for niche audiences.
This Word document by the author has screenshots and better formatting:
http://www.ics.uci.edu/~pazzani/Personalization/ali-kdd04.doc
Posted by: Andy Baio | February 17, 2005 at 10:31 PM
Then why do suggestions still suck so bad? I swear my tivo thinks the only shows I want to watch are kids cartoons and *EVERY SINGLE SHOW* on the Outdoor Life Network.
I used to go through my Suggestions List and rate them but it seems to make no difference so I stopped.
The best way I find shows that I might like? Word of mouth and network promos.
Posted by: gmon | February 18, 2005 at 06:00 AM
Someone should develop a word of mouth (driven by friends) recommendation app / engine using TiVo's SDK. Might have to wait until more functionality is included in the SDK...Should be similar to Netflixes system...
Posted by: New York | February 18, 2005 at 08:10 AM
This post made me think - how much of Tivo's revenue comes from selling rating information to content providers/networks?
There is this long standing debate as to whether Neilsen's diaries (soon to move to set top devices) really work, vs. Tivo's direct connection - plus heuristics to know whether someone is actually watching shows (fast forwards, etc.).
There has to be some action there - Tivo's customer analytics has to be superior (or at least the raw data does).
Posted by: peBird | February 18, 2005 at 12:31 PM
This suggestion system sucks. The previous system which they used until I think about a year ago was much better, giving each box individual suggestions instead of lumping each person together with others with supposedly similar tastes.
Posted by: wombat | February 18, 2005 at 11:40 PM
Finding new shows to watch is the number one thing I find lacking, IMO, to going totally PVR. Without the trailers between shows, it's difficult to find new shows I would like.
And like a few posters but a bit more bluntly, the TiVo suggestion system doesn't do a good job of this. A system that truly suggests similar titles instead of just recording the shows you have put a thumbs up or two on is what is needed. TiVo should be recording what is NOT thumbs up....
Posted by: jrh | February 19, 2005 at 11:37 AM
the word doc is no longer online. does anyone have a copy they could share?
Posted by: andyknas | June 27, 2005 at 08:01 AM
You can get the Word doc from http://ai.rightnow.com/colloquium/papers/tivo_recommender.doc
Posted by: captainsemtex | July 05, 2005 at 06:01 PM
Awww...let's play nice now! Todd Lokken.
Posted by: Todd Lokken | September 23, 2005 at 07:51 AM