Netflix - Not bad! Qwikster - uh, never mind...

(xpost) Haha yes I have.

Barry Lyndon
Eyes Wide Shut
Beetlejuice
The Last Temptation of Christ
Crash
It's a Wonderful Life
Rushmore
Million Dollar Baby

(The other night I got "The Thief of Bagdad", which I know no video store in 100 miles stocks and it was GREAT so I still love you, Netflix. But please don't take advantage of me any more.)

I'm tend just overrate movies that I likeso I get more movies recommendations like those. I also underrate "popular" movies that I didn't like so I get no more movies like that. It's a good system although lately they seem to be recommending more Mystery! series than I would like.

"I vow not to become involved, except when I become involved."

So true. But the pic is broken but I see not what is being linked.

;-O

fuck a very long wait...i have the bitch queued since a release date was announced! technology should know my desire.

currently on a free trial at canflix. it sucks but it's free.

i like my friend list ok - sometimes i get some ideas that way

Anyway, friend me too at this address.

Here are some movies your friends like much more than most Netflix customers:

Dumbo
Edward Scissorhands
Animal House

screenselect.co.uk is pretty decent, and i think they do a free month trial.

And if You Liked the Movie, a Netflix Contest May Reward You Handsomely

By KATIE HAFNER
Published: October 2, 2006

Netflix, the popular online movie rental service, is planning to award $1 million to the first person who can improve the accuracy of movie recommendations based on personal preferences.

To win the prize, which is to be announced today, a contestant will have to devise a system that is more accurate than the company’s current recommendation system by at least 10 percent. And to improve the quality of research, Netflix is making available to the public 100 million of its customers’ movie ratings, a database the company says is the largest of its kind ever released.

Recommendation systems, also known as collaborative filtering systems, try to predict whether a customer will like a movie, book or piece of music by comparing his or her past preferences to those of other people with similar tastes. Such systems will look at, say, the last 10 books, movies or songs a customer has rated highly and try to extrapolate an 11th.

Computer scientists say that after years of steady progress in this field, there has been a slowdown — which is what Netflix executives say prompted them to offer the problem to a wide audience for solution.

“If we knew how to do it, we’d have already done it,” said Reed Hastings, chief executive of Netflix, based in Los Gatos, Calif. “And we’re pretty darn good at this now. We’ve been doing it a long time.”

Nobody with the company will be eligible to compete, Netflix said, so that it does not appear that the contest favors insiders.

James Bennett, the vice president for recommendation systems, said the company had taken great pains to preserve the anonymity of the 100 million movie ratings it was making available to researchers, even consulting with privacy experts to make sure that the ratings could not be traced to individual Netflix customers.

“The data set is the big deal here,” Mr. Bennett said.

Netflix has already used its data set to test the accuracy of its existing recommendation system, so it will be able to gauge the accuracy of each entrant’s set of predictions, executives said.

Mr. Hastings said he thought it was important to make the ratings database widely available. “Unless you work at Microsoft research or Yahoo research or for Jim Bennett here at Netflix, you won’t have access to a large data set,” he said. “The beauty of the Netflix prize is you can be a mathematician in Romania or a statistician in Taiwan, and you could be the winner.”

John Riedl, a professor of computer science at the University of Minnesota and a pioneer in the field of collaborative filtering, said that Netflix and Amazon now had the most advanced recommendation systems.

“Most of the easy stuff has been squeezed out already,” he said, adding that it had become increasingly difficult to make substantial progress in predicting accuracy.

“Any time you start working on any of these scientific or engineering problems, there’s a period of dramatic improvement,” Professor Riedl said. “It slows down because in a sense you’re competing with 15 years of really smart people banging away at the problem.”

Until now, researchers who have been working to improve recommendation systems have been relying on a much smaller database, a set of one million ratings generated by a Web site called MovieLens, Professor Riedl said. “Having a big data set would be really, really useful,” he said.

Francisco Martin, the chief executive of Mystrands.com, a company in Corvallis, Ore., that is developing a recommendation engine based on what people listen to on iTunes, agreed, saying, “With ratings-based systems, you need to rate everything you see in order to get reasonably accurate recommendations.”

Cash prizes in other difficult technical areas have been offered in recent years. In 2004, there was the $10 million Ansari X Prize for a reusable spacecraft. The Defense Advanced Research Projects Agency is again running a contest involving robotic vehicles with the first prize $2 million. And NASA is offering prize money ranging from $200,000 to more than $5 million for building equipment including lunar excavators and solar sails — large mirror-based equipment intended to collect solar power and conserve rocket fuel.

Mr. Hastings said the Netflix prize was different from some others in that it required a minimal financial investment to compete. “This will be one of the largest truly open prizes that’s ever been done,” he said. “All you need is a PC and some great insight.” He said Netflix would publish a detailed description of the winning approach.

If no one wins within a year, Netflix will award $50,000 to whoever makes the most progress above a 1 percent improvement, and will award the same amount each year until someone wins the grand prize.

Professor Riedl noted that a big improvement in Netflix’s recommendation system would be a boon to the company’s business. “It could result in a significant rise in sales if the recommendations do a better job of helping people find movies they want to see,” he said.

I'd enter but I've got other things to do plus not a real mathematician? Maybe I should write in anyway with a spec.

Why do I like this Picasso but not that Picasso even though they're from the same period and share the same styles and brushstrokes?

record store genre labels generally tons and tons more effective than movie store genre labels!

1. Find a dozen or more customers whose movie selections over time have had almost no overlap

2. bucket other customers in with them according to who they overlap with the most

3. resample each customers' preferences for bucket fit every 3 months or so, or constantly, whatever

4. recommendations based on past and current preferences of secret "canary" customers determined in step 1. These can also change over time if one starts overlapping with another one too much- resample and find another set whose overlap values are at or below N, then rebucket.

Do this on an ongoing basis, resample and calibrate at least once a quarter.

Also, keep track of what previews are on what DVDs, and evaluate that with queue-add data to see if it has any value.