What is Digg?61 Comments
- friend18, on 10/12/2007, -1/+21Are you sure you don't like Winnie the Pooh, maybe you think it's recomending movies you won't like, but you should watch them, you might actually love them. :)
- bbatsell, on 10/12/2007, -0/+19Found 'em:
http://www.netflixprize.com/ - Jaymoon, on 10/12/2007, -2/+15It recommends movies you rented that are similar to other movies, which are similar to other movies you rented...
I noticed that when I first joined, I rented Finding Nemo while my little cousins were going to be over for the weekend. Other than Finding Nemo, I have never rented anything probably with a rating lower than PG-13...
After that rental, I keep getting recommendations for Pooh DVDs, as well as just about every Disney movie they have available, no matter how many times I rate these movies 1 star.
I agree with Netflix, something needs to change. - inactive, on 10/12/2007, -0/+13Pandora to the rescue.
I use Blockbuster and its recommendations are just insane. The only things I've ordered from them are Babylon 5 and Star Trek Voyager, yet they're suggesting things like The Notebook. - Desolite, on 10/12/2007, -3/+15how about we bet who wins. i'm betting on google.
- effinnovice, on 10/12/2007, -2/+12Talk to the The Music Genome Project guys.
- greenrider, on 10/12/2007, -0/+9For those wondering - here's how you "prove" your system is better:
1. The data set containing user ratings of films is arbitrarily split into two halves - the "training" set and the "trial" set
2. Your algorithm uses the information in the "training" set to build predictions for what's in the other set
3. Your algorithm's predictions are compared against the "trial" set. In other words, you will have the information that "User C rated Waterworld 2/5," while your algorithm will say "On the basis of the training set, I would have predicted User C would have rated Waterworld 3/5."
Thus, performance is measured on the basis of:
a) Standard deviation of the algorithm's predictions vs. the actual ratings in the trial set
b) Time complexity required to run the algorithm on a database of a given size
This all falls under the heading of "collaborative filtering" research. Approaches to building a collaborative filtering algorithm could include something as simple as "suggest any movie that has a high average rating" to more complex approaches such as "find the top 20 customers whose tastes are most similar to this customer, and recommend what they collectively liked"
The difficulty inherent in collaborative filtering is to make the standard deviation as low as possible while keeping the time complexity reasonable (i.e. not O(n^3)) so that it will scale gracefully as more ratings are added. - elroy, on 10/12/2007, -0/+8My plan is just to spit out a static list, whether or not you've already seen them.
1. The Big Lebowski
2. Breakin' 2: Electric Boogaloo
etc. - friend18, on 10/12/2007, -2/+8How does the system work now?
- PradaPete, on 10/12/2007, -2/+7really? I'd have more fun with a girl's *****
- imtigger2, on 10/12/2007, -1/+610 star? that's it!! I'll be right back.... have to contact Netflix now.....
- zonk3r, on 10/12/2007, -3/+8the biggest problem with this contest is proving your system really is better.
- endgame, on 10/12/2007, -5/+10Netflix....ROCKS!
- Acrion, on 10/12/2007, -1/+6If nothing else, they offer a 700mb dataset to play with. That is certainly something that I could have fun with.
- blinkfink182, on 10/12/2007, -1/+5No one would rate Waterworld higher than 0/5
- bbatsell, on 10/12/2007, -0/+4From the contest rules:
"We provide you with a lot of anonymous rating data, and a prediction accuracy bar that is 10% better than what Cinematch can do on the same training data set. *(Accuracy is a measurement of how closely predicted ratings of movies match subsequent actual ratings.)*" - BobTGoon, on 10/12/2007, -0/+4Hmm. I'm pretty impressed with the recommendation system as it is. Would be interested to see what improved system would come up with for me.
- chesterjosiah, on 10/12/2007, -0/+4Where are the details, from Netflix, for the contest?
- jkilby99, on 10/12/2007, -0/+4Jaymoon:
I had something similar happen to me when I first joined Netflix (I rented an animated film or TV show, can't remember at the moment). I'd suggest either:
1) Keep rating movies... eventually, it will move away from those flicks.
2) Whenever it recommends those Olson Twins movies, click "Not Interested." Eventually, it'll get the point that the disc you rented was an outlier in your movie tastes.
From my personal experience, the system works pretty well. It has even managed to recommend movies that I rented and liked or ones that I have been wanting to see but haven't put in my queue. - madtowner11, on 10/12/2007, -2/+6Isn't it 4 stars? Even with 5, that's not quite enough to distinguish a movie you love from a movie you really like. They should make it a 10 star system.
- jkilby99, on 10/12/2007, -0/+3You have to register a team to access the data (which is why there is not a link to the data). The link to the site has already been mentioned: http://www.netflixprize.com
- HyperbolePolice, on 10/12/2007, -0/+3I suspect this is true. My guess is the difference between a 500 line naive implementation and the Netflix system is 10%, and increasing another 10% is impossible without better user data.
- TheKillDoctor, on 10/12/2007, -1/+4You rate movies 1 to 5, the system then recommends movies that it thinks you will like to watch. The more movies you rate the better the recommendations you'll get from their database of movies.
- invader, on 10/12/2007, -0/+3my idea is better than the 10 star idea!
it's.....
NEGATIVE stars!!! you rate it negative, then they destroy the physical DVD! Brilliant!
now i'm off to netflix... oh crap, i shared my secret to winning with digg's entire user base! - shatters, on 10/12/2007, -1/+4but if they rated all movies, with said actor, a 1 of 5, there would be a pattern.
- ragipy, on 10/12/2007, -1/+4yep really, they are not mutually exclusive
- camvet, on 10/12/2007, -0/+2What I can't figure out is when they will give out the prize? Is it the first person to 10% or better gets the prize, or is at 2011 whoever is 10% and the best of the teams is the winner?
- inactive, on 10/12/2007, -1/+3I can't recall that I have ever listened to a recommendation given by a website like google or netflix. I personally think its a waste of money. I know what kinds of movies I like personally and can easily find something I like by searching through the categories.
- TheKillDoctor, on 10/12/2007, -0/+2Just double checked, it is 1 to 5 stars.
Though I'd love to give half stars for some movies... - Pissoff, on 10/12/2007, -0/+2The current system is pretty bad, I've got around 1595 ratings and only 3 or 4 recommendations, of those 3 or 4, I would watch 0, so I mark them not interested, and I now have 0 recommendations. The more you rate the more recommendations you get? False!
- davodavo, on 10/12/2007, -0/+2I know what you mean. I have 495 ratings in all genres and my only recommendations are in foreign films and documentaries - and the ones it has recommended that I've seen have been terrible.
- Pissoff, on 10/12/2007, -0/+2My bad, I have 1755:
Movies You've Rated (1755)
ALL RECOMMENDATIONS
Get more Recommendations by rating more movies.
To get recommendations, use our Movie Rater to rate the movies you've seen and loved. For other recommendations, you can also click the button below to see which movies your fellow Netflix members have rated highly. - displaynone, on 10/12/2007, -0/+2I know, if you like a movie you can, what I like to call, 'digg-it'. Then the users with the most submitted movies can take over Netflix. Muhah ah ah ha ha!
- cl0r0x70, on 10/12/2007, -2/+3Ok. But how do I get the data? There are no links in the article.
- diffusio, on 10/12/2007, -1/+2I think Last.fm's recommendation system is pretty good. They recommend me artists that I like, but haven't listened to on my computer yet (thus scrobbled into their system).
They'd be able to win if they wanted to port their system over to movies/Netflix. - kolobcreek, on 10/12/2007, -0/+1Viva la Bounties
- kirakun, on 10/12/2007, -2/+410% lift is impossible.
- DanDaMan, on 10/12/2007, -0/+1I guess giving Amazon.com a million dollars didn't work. Their recommendation system works 10x better. Or, yeah, get a labor-intensive system like "The Music Genome Project".
- tdeletto, on 10/12/2007, -1/+2This exists and it was done years ago...and I'm sure they know it.
There was a site moviecritc.com that was the test application for a company...what was it called...LikeMinds...anyway, the company was purchased by Macromedia, so it is now owned by Adobe.
It used collaborative filtering (an awesomely powerful thing, still vastly underused on the internet) and produced remarkably accurate predictions (once you had rated a good sampling of movies).
I don't know what Netflix uses today, but clearly collaborative filtering is the way to go. What they need is more detailed ratings, but at the same time they also don't want it to be difficult to enter ratings.
The closest things I know of today are a site called movielens (a university research thing, I believe) which is pretty good, but not quite as good as I recall moviecritic.com being and moviepig.com, which despite its terrible name, has an awful user interface. - white2grey, on 10/12/2007, -0/+1I think they should compare one user's list of rankings with the other users and find user's who have similar tastes... then they could recommend movies from the other users' lists that the original user hasn't rated.
- kuya, on 10/12/2007, -1/+1Netflix: Powered by Google.
- kuya, on 10/12/2007, -1/+1or Netflix: Powered by Yahoo Movies!
- evansls, on 10/12/2007, -1/+15 Years is a long time to develop the next gen algorithm for movie choice filtering. Star rating don't work, because stars don't represent anything. Sure 5 stars means PERFECT MOVIE OF ALL TIME and 1 star means THE WORST MOVIE OF ALL TIME... who cares? By looking at the numbered stars tells you nothing about the movie. I have a better algorithm idea, but since I'm not a programmer i can't build it. I am a designer and can build great story boards too... so, I'm going to write out my idea to see if anyone out there can build it. I could care less about the grand prize money, because if I was able to program my idea I would sell it for a lot more than 1 million dollars.
- tylerdurdenclub, on 10/12/2007, -1/+1yeah, it is already 95% accurate so getting another 10% is going to be very hard.
- ultraboy13, on 10/12/2007, -1/+0This one works good already.
http://like-i-like.org (movie recommendation system)
Have you ever dreamt about watching only the best movies? Figure out in advance which movies you'll love and which you'll hate. Don't waste your time any more. %) - dclowd9901, on 10/12/2007, -2/+1I think they're looking for something more original. Elegant and brilliant. Or they're just putting on a gimmick.
- Suplyndmnd, on 10/12/2007, -4/+3Lemz, that's exactly what they can do. A million dollars is a million dollars but no way in hell would I take a chance of sinking a year or so into writing code to get that 9% and them just use my code later saying "Oh, we said we'd pay if it improved it 10%, you only got it to 9. But hey, thanks for the code"
I think my next plan would be to exterminate NetFlix at that point and that i'd do for free. I think this is just a gimmick and I seriously doubt anyone will A.) Win or B.) Waste their time. Gimmie a lifetime subscription (10 at a time) and i'm all over it though lol - FackBlog, on 10/12/2007, -4/+2I wonder if this is a genuine attempt to improve their system or just a publicity stunt to show how great it currently is .. how would one measure the accuracy of something as subjective as "liking a movie"?
- GuyHitByTruck, on 10/12/2007, -3/+1You don't get out much, do you?
- GuyHitByTruck, on 10/12/2007, -3/+1I was thinking the exact same thing. Not surprisingly, you've had seven responces and no answers. I'd imagine it'd be pretty difficult to fix something you know nothing about.
-
Show 51 - 60 of 60 discussions
Check out the new & improved 
© Digg Inc. 2009
— Content created and posted by Digg users is