BT et C

Tuesday, July 19, 2005


oh dear oh dear this is dangerous. I don't picture myself ever darkening the door of a Blockbuster again. I am a netflix junkie.

Sounds annoying at first: when you want a movie, you want it now, don't you? You don't want to wait for it to arrive in the mail. In actual practice, you don't really notice the wait. Kudos to:
-whoever came up with the idea of the queue ... I put twentysomething DVDs in mine after about 15 minutes of searching
-whoever designed the peel-and-fold-and-stick return envelopes. It seems like such a little point, but it matters. If returning the disc were even the slightest bit unwieldy, the whole business could collapse. I'm not joking.

Like just about EVERY "recommendation system" on the net, this one is dumb as a post. This is because computers are dumb as posts (except for the ones that run Gentoo). Example:

I tell Netflix that I liked the movie "Hero". A lot. This does not mean that I want to watch every Jet Li (or Bruce Lee ?!?!?) movie ever made. What's missing is some communication about why I like "Hero". If I could get that across, even the posts could figure out that I'd like "Raise the Red Lantern" more than "Fists of Fury 3".

Well, this is an example (like REST) where you gotta make do I guess. Improving the intelligence of the algorithm probably requires a serious increase in how much your users have to contribute*, and (like the peel-and-fold-and-stick envelopes) users don't contribute %^##% unless you make it staggeringly easy.

*e.g.: you'd need an Allocation-style assessment of the film/music/whatev instead of a 1-to-5 star rating system.
Acting:          200 ------------- 1000
Writing:        200 ------------- 1000
Production   200 ------------- 1000
Genre           200 ------------- 1000

Maybe the (relatively few) users who are opinionated enough to do this would be sufficient...


  • I propose an alternative means of relating the "why"...

    use tagging like technorati and flickr.

    so you can rate Hero high, like a 9 out of 10, and then tag it with "visuals", "writing", and "ass-kicking". then netflix (or, which I use) can look thru all the users that have given Hero a high rating AND have tagged it similarly to your tags. it will then search those users' rentals for other high ratings with similar tags and present them to you.

    so the list you see is a list of other similar movies from people who liked Hero for the same reasons you liked Hero. you could even tell it to find non-similar movies - movies that are not necessarily like Hero, but are from people who liked Hero for the same reasons you did.

    By Blogger luke, at 7:12 AM  

  • it could even be a possible web service app that does it. then you open up the API's to allow input from netflix,, and Amazon feedback. charge the major sites some kind of monthly fee, get some hefty servers and you're good to go...?

    By Blogger luke, at 6:21 AM  

  • Hmmm... don't know if you'll get way down here in the comments, but this is probably "correct". I'm pretty interested in the thingamy apprach, too. Big poetic post on that stuff coming later this month probably.

    By Blogger Matt Crouch, at 4:23 AM  

Post a Comment

<< Home