Slashdot Mirror
Build a Better Netflix, Win a Million Dollars?
An anonymous reader writes "In a quest to better movie recommendations, Netflix is opening their database (nytimes, registration and first child required) to users to try to craft a better recommendation technology. The problem is not easy. Says one researcher: 'You're competing with 15 years of really smart people banging away at the problem.'" Recommender systems are really an interesting problem, and that is likely very interesting data to play with.
"Nothing for you to see here. Please move along."
Have they now?
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.
;)
But if someone does win within a year they will still have the ability to use others' code, free of charge, as part of their product.
The article doesn't say but how will you know if your code is making choices better than their existing system? I wouldn't be submitting my code unless I was sure I was going to win. Then again I'm not a gambler or a coder
I officially announce I will be entering BigAtticHouse's Vectorspace Database into the melee. At least to see what might come of it.
meh
So, the professionals have been working at it for a long time. Is it safe to assume some teenage to early college hacker will find a success within two weeks.
34486853790
Connection too slow for X forwarding? Try "ssh -CX user@host"
if(user.getGender()==Person.MALE)
recomendation=MovieGenre.PORN;
else
recomendation=MovieGenre.CHICKFLICK;
And of course, slashdot must have sensed my post as my image word is "pervert"
A lot of ppl is going to waste mucho time on this one. I recommend watching the movie Syriana instead.
..except, instead of making it open to the community (which is not a bad idea, I must say) I thought of having Google do it. This is, perhaps, IMHO, a much better idea. Now, what we really need is a Movie Genome Project, much like the Music Genome project that lead to Pandora.
Zagreus sits inside your head, Zagreus lives among the dead, Zagreus sees you in your bed and eats you in your sleep.
They have decent tech for building similar/recommended alternative pages.
Especially the newer blogish type pages where theres a gallery and a small selection underneath.
Not that I would know of course.
liqbase
As a NetFlix user I have one suggestion for their recommendation system that can make it much better. Make it aware of the connection between series. That is to say, If you rent season 1 of something, suggest season 2, not season 4 (even if season 4 has better review ratings). If I mark season 1 of something as "not interested" instead of giving it a user rating, don't suggest every other season of that same show at the top of my recommendations. I mean how many times do I have to tell you I don't want to see any season of "Friends" ever, even if you pay me?
How will they handle privacy issues? Don't the same issues appear here that appeared with the AOL data this summer? With enough ratings you can narrow down to a specific person, and then find out about all the pr0n that this person has been getting as well.
Why is it that the Slashdot editors are just too damn lazy to look up the RSS feed links to these pages?
While this may be true, I wouldn't let it deter you. Collaborative filtering is a field that is far from dead. The interesting thing about collaborative filtering is that on the surface, it seems pretty straight forward but once you dig into the mechanics of it, there is actually a lot of playing you can do. Ironically, the way you display the data to the end user is often what determines how well of a job you did.
Allow me to take a naïve approach at this topic and say we generate a movie index of each person. I would have A Clockwork Orange and Koyaanisqatsi at 5 while The Ring 2 would be at the very low end. My friend might have similar movies. If he has A Clockwork Orange up there, you might be able to compute a Euclidean distance between us. However, this approach falls apart because no one has seen Koyaanisqatsi and of the 20 movies I've ranked highly, they are hard to find.
You don't have to stop there, however. You could also database the movies I marked as "uninterested" or the movies that were presented to me but I didn't vote on. Like if I had seen the offer to mark J-Lo's latest flop but didn't, wouldn't that tell you something about me?
So these caveats present themselves all along the way and, at the end computation, you have many different strategies for this data. For example, while you might not be able to link my friend an I through movies, how far apart are we on a nod network? What I mean is, if you plotted every user in their own dimension depending on the movies they ranked and attempted to compute as good a distance as possible between all users, how far would I be away from my friend by hopping on these nodes? There's a lot of information to be gleaned in this sort of friend-of-a-friend collaborative approach.
Now you need to present this information to the user. Do you just up and recommend him a movie? Do you take Amazon's approach and say "Other people did this -- so should you."? Or do you give them some sort of three dimensional flash plotting of you versus the people nearest to you? Do you allow the user to contact those closest to them? Those farthest away?
My point is that while 15 years of research has been done, it doesn't mean there's been 15 years of testing and implementation which, in the end of creating products, is where most of the importance lies.
My work here is dung.
Link everyone's credit report into their movie preferences; I'll bet your complete credit history would give them a 5-10% better chance of picking your movies. But seriously...why isn't this just a regression exercise?
According to the article, if no one wins within a year, Netflix will award $50,000 to whoever makes the most progress above a 1 percent improvement, every year, until someone wins the grand prize.
--
Go where the Web Thinkers gather
Watch the Teaser Trailer for "The Lightning Thief" Her
Constructing a perfect recommendation system is easy.
I have discovered a truly marvelous demonstration of this proposition that this post is too narrow to contain.
The problem with recommendation systems is that they use too little information to catagorize their subject.
What they need to do is copy the methods of the Music Genome Project (www.pandora.com), and list a larger set of attributes for the films. This way it can recommend films by checking many more characteristics, such as director, tone, writer, or subject.
Recommend Death to Smoochy and Date Movie, please! I hope Richard Stallman enters the contest and wins. I bet he would make good use of the one million dollar prize.
ALTERNATIVE FREEDOM
A documentary about the invisible war on culture.
Features RMS, Danger Mouse (of Gnarls Barkley and the Grey Album), Lawrence Lessig, and more...
http://alternativefreedom.org/
If you can beat "15 years of really smart people", then your work product probably has more than a million dollars in value if you were to license it out to places like Amazon, eBay, Netflix, etc. Even a 1% improvement in revenues from a 1% improvement in recommendation accuracy is probably worth more than 50K, if sold to the major e-tailers. On the other hand, if you just want an interesting problem to screw around with in your spare time and don't want to go through the bother of forming a company in order to monetize that work, this is a pretty cool opportunity.
News for Geeks in Austin, TX
I wish they'd fix the problems in the logic determining what they actually send me from my queue before fixing problems with what they recommend to me. If I've got season 1 of a show in my queue prior to season 2, don't start sending me season 2 because some disc of season 1 is unavailable (which has happened to me multiple with both netflix and blockbuster online), send me something else completely. They've got the tech to keep one season of a tv show in order, it can't possibly be that difficult to extend that to keeping multiple seasons of a show in order.
On top of that, don't show me that it's available in my queue but send me something else instead. While I haven't asked netflix about this, I have asked blockbuster online, and I imagine they are both doing the same thing. The disc is "available" just not at the warehouse used to ship to me personally. Instead of basing one piece of information off of total stock and one off of local stock, base them both on the stock at the warehouse shipping to me.
You can trick the NY Times personally but you can't do it from a front page of a widely popular commercial site.
I think it is the reason.
Slashdot can't send thousands of users with a fake referrer to NY Times. That link you provided is for people using RSS readers and subscribed to NY Times RSS feed.
I think they should talk with NY Times web team to allow slashdot readers with referrer=slashdot without needing login. They can arrange it for sure, this isn't a "no name" site.
It would be nice for NY Times for statistics too. I bet they currently have to tweak the statistics for "fake" RSS links from Slashdot.
About "no ads" version: It would be like NY Times mentioning Slashdot and sending people to some other domain (slashdot sux? I forgot) which doesn't have Slashdot ads which makes this site work/pay for the costs. That also means hundreds of thousands users.
I am not apologising for NY Times or trying to start a discussion about advertising, I just say my end user point of view and plain guesses.
If they wanted users to rent more DVDs, they should stop throttling them first...
http://www.msnbc.msn.com/id/11262292/
Now some 12-year old will come up with a great systems and NetFlix will rape them for theor code and call it a day...
http://www.kontentdesign.com/
If Netflix doesn't have the movie in stock it should burn the movie on demand.
> ...Netflix is opening their database (nytimes, registration and first child required)...
In order to view the article sacrifice your first born on the AJAX altar to the right. Use drag-n-drop pentagrams as necessary.
My UID is prime. Hah!
Any marketer will tell you that what people tell you they want and what people actually want are very different things. Even if people answer honestly, the data you gather is often unreliable: people simply don't have as good a handle on what they want as they think they do.
Not that marketers have a better handle, but simply that people will swear up and down that they would buy a peanut-butter-filled hot dog, that they loved the one they tried, and then don't actually buy any.
Don't believe me? Go see Snakes on a Plane. Nobody else did. (Sure, $33 million seems like a lot, but that's chump change for a major studio release these days.)
The best improvements will come from insights gained between the lines. You may have rated The English Patient eleventeen stars, but if your next seven rentals were all episodes of The Girls Next Door, which you only rated 3 stars, it certainly looks like you want more Hugh Hefner and less Ralph Fiennes.
The best data is the data that the subject doesn't realize he's giving you. Once you start imposing conscious choice on the ratings, you get only what they say they like, not what they really like.
I stopped rating movies after I found that I got recommended a lot of crap. Say I rent a slasher movie that, for its genre, is artfully done. I rate it high. Now I have recommendations for a bunch of worthless, straight-to-video stuff that I really don't want to see.
This is the real nut to crack, IMO. How do come up with an algorithm that rates 'quality,' an elusive concept that means different things to different people?
Not to mention, I'm fickle.
Dark Reflection
I personally weigh movies on a number of different factors. I might give 3 stars to a movie because it has 4 of my favorite actors in it even if I didn't care for the plot. I might give 3 stars to a different movie with horrible acting but interesting camera angles (From Dusk Til Dawn 2). I tend to average out my ratings dependent on many things a movie has to offer.
The problem is is that that is my rating system. It works for me. But it does little good to anybody else because they are rating based purely on something else.
I think they need to implement the ability to rate more aspects of the movie. I'm sure some people out there rate the movie poorly if their disc is scratched or the transfer quality is poor even. A simple 1 to 5 system doesn't cut it. People rate things that aren't "Was the (romance) plot good?", "Do you like this director?", "Do you like these actors?". People rate things that aren't on the box.
If I had a way of greatly improving the current Netflix system, I wouldn't bother with the prize; it'd start up a competitor. If it's that much better than Netflix, it'll be worth way more than $1 million.
And some macaroni pieces.
Yes, they need more characteristics of movies.
But they also need ways to identify the characteristics of people's choices. Right now, one NetFlix account can be used by a whole family. So instead of getting 1 person's characteristic choices (teenage emo goth girl), you get those combined with the other family members (Dad's action films, Mom's chick flicks, Jr's teenage sex comedies).
Eventually, you'd end up with a movie genome cross indexed to a sub-culture.
SELECT TOP 10 title
FROM tblMovies as m, tblAdvertisers as a
WHERE m.studio = a.studio
ORDER BY a.adRevenue DESC
I win.
Just a simple suggestion. A lot of the DVD's that NetFlix recommends I already own. So I can't put "not interested" because it'll filter out those types of movies for me. How about something like "I own it" as an option, and it can pick recommendations based on those as well?
Does anyone know how to enter the contest?
As cunning as a fox, which has just been appointed professor of cunning at Oxford University. http://www.kinlan.co
Encourage end-user tagging, compare on popular tags for matching a la LJ "people who have the most in common" search.
Or leech off of IMDB's recommendation system, which seems to be quite good.
Terrorists can attack freedom, but only Congress can destroy it.
Yahoo has just done it and now netflix is, these companies are not hiring programmers anymore but asking hobbyists to write them code in their spare time and the hobbyist with the best code wins. Which really eliminates the need for programmers.
I rekkon this will bite us all in the ass in the future because if everyone does this there will be no jobs left and therefore people will not learn because there is no money involved and then the companies will have no hobby programmers left.
- Collect more data from the person: age, region ("The South" "Northeast" etc.), education level, etc. As far as privacy goes, if you don't want to enter it that's fine--but your recommendations will suffer
- Analyze rental activity with ratings. If I've rented a movie 3 times, I probably like it more than the movie I've never rented but gave 5 stars
- Analyze queue transactions. What movies did I add to my queue together? What did I move up or down? What movies did I delete from the queue (and ask why: saw it already, changed my mind, etc.)
- Analyze how long I hold movies versus what I rate them. Upon return ask if I watched the movie or not.
- Find more ways to group movies together (genre, subgenre, actor, theme, director, writer, etc.). Figure out which actors I love/hate
Bigtime Consulting - "We're the best because we cost the most"
"The problem is is that that is my rating system."
I re-read this 5 times and then gave up!!
Opinion:=TMyOpinion.Create(Me);
BitTorrent!!!
All jobs work this way (or at least they should). Return on Investment. In order for a company to make money, they will pay you a wage. Hopefully you will produce work that is at minimum equivalent to the amount that they pay you. If not, then they will be losing money employing you and if they have decent reporting/management will probably fire you.
Many companies do offer incentive programs (more likely for upper level positions), but is still just a percentage of the actual "value" that you created, not all of it.
Sidenote. We all like to complain about how overpaid execs are. They usually don't have earth shattering ideas. However, if they find a way to increase efficiency by 10% but it is for a $100 million dollar project, then they essentially create $10 million worth of "value." If they only get a 10% cut, then they get a paycheck of $1 million, even if they didn't do any of the work to actually realize the improved efficiency.
When I have a kid, I want to put him in one of those strollers for twins and then run around the mall looking frantic.
Disclaimer: I subscribe to the same sort of service, except through blockbuster... maybe Netflix does have this feature. My wife and I share a queue... I imagine many, many of these queues are shared. We have very, very different tastes in movies. Instead of getting recommendations that suit us both (which is next to impossible), the recommendations just get very, very confused. If I could just keep my and her recommendations from tangling, we would both have an easier time.
I see that the NYT article linked to just about everything except MovieLens. I've used the site, and folks might like to try it out. It looks simple, but it's fairly nice, having some of those fun dynamic pages that are all the rage these days. One neat thing in comparison to Netflix is that it will give a projected star rating for you, rather than simply saying "Recommended".
Of course, I'm biased since I had John Riedl as a professor in a few easy classes. I think he tried to spin off this research as a new company, but I'm not sure if it ever got off the ground.
One thing I'd really like to see has little to do with the quality of ratings, though. I'd like to be able to keep a common database of my ratings across multiple sites. At the moment, I've rated a number of movies at Netflix, MovieLens, and IMDb, but they aren't entirely consistent. Unfortunately, two of the sites use a ten-point system (IMDb has a ten-point scale, MovieLens goes up to 5 stars, but in half-star increments), while the other uses a five-point one (maybe six if you say "Not Interested"..).
Well, I'll have to poke around a bit with this stuff. I wouldn't be able to do much, though, since my level of knowledge in this arena is very limited...
Netflix allows you to have up to five profiles, each with separate queues. According to the site, "Each profile has its own ratings, recommendations, Friends and MPAA levels."
Slightly disreputable, albeit gregarious
Residents of the province of Quebec in Canada are ineligible to participate. :(
So now developers should gamble their knowledge and skills at this in hopes of the payoff and publicity instead of getting a contract for 1/100th of that to accomplish the same objective over likely the same time frame? Hmmm....let me consider that one for a while...
Just another nameless binary in a crowd of 1's and 0's
If you truly and honestly believe that the winner of these contests are not then promptly offered jobs, you are missing the point fo the contests. These contests are generally more glorified job interviews then anything else. I doubt they even expect anyone to win. What they DO expect is for people to send in some innovative solutions. They will then go out and try like hell to hire the people with the best submissions.
If you are looking for a job, I wouldn't view this as a competition to make you obsolete. This is a competition to find a new employee and offer him a sweet sign on bonus.
Disclaimer: I subscribe to the same sort of service, except through blockbuster... maybe Netflix does have this feature. My wife and I share a queue... I imagine many, many of these queues are shared. We have very, very different tastes in movies. Instead of getting recommendations that suit us both (which is next to impossible), the recommendations just get very, very confused. If I could just keep my and her recommendations from tangling, we would both have an easier time.
This problem is already solved.
With Netflix you can have multiple queues (up to one per disc-at-a-time out) and reassign the "number of discs out per queue" from 0-#out as long as the total isn't greater than #out. It also handles reassignment with discs outstanding well.
The result in my family is that we end up with independant queues and independant recommendations. If Blockbuster offers the same feature you could split your queue up and get what you want. On top of that you won't have to keep organizing your queue to get the correct movie next if someone takes time to get around to watching their movie.
There are two kinds of people: 1) those that need closure
They had a large property where they just couldn't find gold. They made all of their geology data public and then held a $575k contest for the best suggestions. Using this Open (Gold)Source technique, they found numerous deposits and dramatically increased their market capitalization.
j html?articleID=159907864
http://www.intelligententerprise.com/showArticle.
Maybe the problem with Netflix is good old classification. For some reason, the movie industry feels compelled to call everything "Action" "Drama" or "Comedy" -- add some tagging, use the Dewey Decimal System, I don't care, but make an effort at some descriptive categorization and see if that increases repeat buys.
I just can't trust a user rating without reading their anecdotal description -- I want to know why something was rated only 2 stars before I pass judgement. Maybe Sally was a first-time buyer and decided to rate everything she rented as 1-star because they double-billed her credit card. Who knows? And Personally, I never rate stuff. I just don't care and can't be bothered; I'm sure I'm not alone either.
The software security experts always say, "never trust user data" -- maybe this applies to recommendations as much as SQL injections.
body massage!
# Netflix has more than 1 billion movie ratings from customers. The average subscriber has rated more than 200 movies.
# Netflix members select approximately 60 percent of their movies based on movie recommendations tailored to their individual tastes.
# Netflix's members rent more than 95 percent of all titles in the Netflix library each quarter.
http://web.netflix.com/MediaCenter?id=1005&hnjr=8
Wincopy
Here's hoping somebody can fix the current system, because NetFlix and I are at a stalemate. With 1,324 movies rated, NetFlix gives me 0 recommendations because it has become quite confused about my taste in films.
Once, I rented and liked a Devo DVD, so it recommended every band with a concert movie, but I don't like every band and started marking things "Not interested". Then, I added a Sarah Silverman disc to my queue, which NetFlix took to mean that I love all stand-up comics, especially those on the Blue Collar Comedy Tour. But I'm not really interested in all comics, so I had to mark them "Not interested". Then I checked out a few documentaries, sending NetFlix into an excited state where it recommended all kinds of uninteresting docudramas. "Not interested".
At this point, NetFlix thinks I'm not interested in anything even though I have a number of films rated 3 stars or greater.
But maybe its recommendations are right... Maybe Sarah Silverman and Bill Engvall really are for the same crowd.
how about instead of sending out DVDs that have been scratched to death and unplayable you actually inspect the discs when people send them back in and charge them accordingly if they have damaged the disk to the point that it needs to be replaced?
The movie industry and netflix have an obvious solution to increase sales and interest and push more disks and also cut down piracy, and they have always had it, use advanced tecnology as it becomes available. DROP the price down to the rental fee (around there, very close anyway) and let the movie going public have the disk. Let them keep the thing. They could do it too, stamped disks are cheap, any number of "alternative" businessmen who operate their own "courtesy disk distribution channels" show what stamped disks cost in bulk with some data bits on them. A buck or two max beyond what those gents charge would be a good thing for all concerned. Volume sales, a time honored concept to "make profit". We have the technology to do this, this is one place where we have sci-fi coming true-we have cheap "replicators", I sugggest we adopt that technology across the board.
;)
For netflix that would mean one way shipping, a significant cost savings, for the outside shipping fees and halving the labor handling.. For theater owners,this a great way to get people to watch the movie the first time at the big screen theater and buy their lard brand popcorn and ~cola~, a complimentary disk of the movie the people just paid cash to see on the way out.
The next step is drop the number of movies produced and do some more work on screenplays and plot, etc. Just make better movies, but less of them, right now, just too many out there, dilutes the pool a lot. Then people might actually go and *enjoy**** the movies, and tell-a-friend about them.
*****this still leaves annoying cellphone ringtones in the movies. Not that I would suggest anything illegal, immoral or fattening, but if doofuses who don't turn their phones off got a 5 gallon or so ~cola~ and lard brand popcorn shower from their surrounding movie watching neighbors in the theater every time a ringtone was heard, the problem might become self correcting. IANAL, check local vigilante regulations first
I share a netflix account with 4 others in my family. We all have very different tastes in movies. I like the horror and foreign film genre, whereas my brother prefers oldies, and comedies, while others are 'Friends' fans. An algorithm would need to keep this scenario into consideration as well, as I am sure this is not uncommon.
:n
Did anyone realize that Netflix is releasing 100 million of anonymous customer data? I thought
Netflix also needs to introduce a dvd iso download service. In an age where most people own dvd burners, Why the hell not open dvd iso download service?
\
Well, if you think about it objectively, you find that genre is a terrible way to recommend movies. Consider:
Movie A is an artful horror film (we'll even give it a bonus point for originality.) Movie B is a low budget straight to video rehash of Movie A.
What are the differences between the two?
A) Budget / production values.
B) Production company.
C) Actors.
D) Crew, particularly director / writer / producer.
E) Originality of script. (This is kind of subjective, but surely a remake of a movie is less original than the original.)
These are good criteria for picking out films. It's up for you (and every other user) to tell you what the values of these mean.
Genre is terrible for choosing a movie; it is much better suited for discriminating out other movies. Someone can filter out all romantic comedies, fine. But saying all romantic comedies are alike because they are romantic comedies is wrongheaded.
If I both liked Shrek II and The Motorcycle Diaries and Maria Full of Grace, and you liked Shrek II, does this mean you'll like the other 2? Absolutely not necessarily. The problem with the recommendation system is the limitation in -expression- of rating. A 1 to 5 star scale for every movie? It will never work.
Combine tagging with rating and you'll find a much better recommendation system.
i.e.: "Shrek II: cartoon, comedy, satire, Mike Myers; 4.5 stars" vs. "Shrek II: 4.5 stars" and "The Motorcycle Diaries: documentary-like, Che Guevara, cinematography-driven, story-driven, soundtrack-driven; 5 stars" vs. "The Motorcycle Diaries: 5 stars".
Then you start to do more than correllate which films are well-liked and which ones aren't. You start to correllate sense of humor, style, etc.
On iTunes, one of the better features for me is the iMix. Let's say I like some Mason Jennings song. It shows up on a few iMixes, so I check out those iMixes and find songs I like, be it Jose Gonzalez, Teitur, Alexi Murdoch, Patrick Park, whomever. It beats the crap out of Apple's "Just for You" recommendations, which are skewed to hell because I've bought some Madonna songs (and maybe a few others) that are simply -too popular- and completely mess up the bulk of the recommendations because of the sheer number of connections those few songs have.
MORTAR COMBAT!
I agree. Similarly, if I rent something to watch w/ someone else, namely some chickflick my girlfriend begged me to watch w/ her, I am now forever getting chickflicks recommended to me. I see this a good deal more w/ Amazon- I buy people gifts off of there, and now Amazon thinks I am really into cooking. I bought a service manual for my car, and now Amazon thinks I am a car buff. Similarly, just because I didn't like "stupid cash-in sequel" 3 of a series, doesn't mean that I dislike the entire series, which is often how it seems to work.
This in particular seems fairly easy to fix though- Only recommend movies based on trends, not single data points. If I start renting tons of chickflicks, well then... I have what is coming to me.
I am seriously excited about this, but it sounds too good to be true? Please point out any traps in this competition?
The winning conditions seem fair to me, as far as I understand, I even get to keep the rights to my algorithm, I "only" have to give a free license to Netflix? Or are there any traps I don't understand?
And look where it got him. In the nuthouse.
Followed by a multi-million dollar publishing career. Ok. Good point.
-p
My problem is that I rate the movies I watch very high. Why? Because if they were crappy movies I wouldn't have rented them anyway! I usually know something about a movie before I rent it, even if it's just the viewer reviews on NetFlix. My ratings are 3, 4 or 5, only very rarely do I give out a 1 or 2.
Consequently NetFlix thinks I like everything. While the system is smart enough to not recommend Martin Lawrence movies, it usually gives me movies I'm simply not interested in. Or at the other extreme, it gives me excellent movies that I've already seen, just not rented at NetFlix. For example, right now it is recommending:
Anne of Green Gables - Ugh
Eddie Izzard: Dress to Kill - WTF?
Eleanor & Franklin: The Early Years - I'm falling asleep...
The 39 Steps - Good choice, too bad I've seen it 39 times
Don't blame me, I didn't vote for either of them!
This is an easy problem. Simply write a predictor algorithm to compare the affinity characteristics of a given film to the affintiy characterisitcs of a given subscriber. Based on the goodness of fit, probabilities of acceptance can be assinged and recommandations made. "You like Romances more than Comedies, and both of them more than Westerns? OK, based on the weights you've given, High Noon isn't for you, but Young Frankenstein might be better."
One problem, how many dimensions ARE there to human affinity (eHarmony thinks they know)? And how do they interact? I like Romances, and Comedies, but NOT Romantic Comedies, unles they ARE in a Western, in which case they are about the same as an Action - War - Drama for me.
And by the way, can you infer these things from my watching habits, or do you have to ask me, and what if I lie (or just don't know/realize I'll like something? Or maybe I just rented Silverado (Western Action Romance Drama) because I happended to know one of the producers, and I can't stand Westerns otherwise?
And of course it all goes out the window when it turns out I like Samurai films, and Magnificent Seven is a remake of the Seven Samurai.
Piece of cake. How hard can it be?
Behold, this dreamer cometh. Come now, and let us slay him... and we shall see what will become of his dreams.
I certainly hope Amazon chooses to license anything that comes out of this. I've been a Amazon customer for about 10 years and have bought a couple of hundred books from them. For the first 2 or 3 years they gave me pretty good recomendations and I found a number of new authors that I probably wouldn't have started reading. Over the last few years they never catch a new author and suggest them.
Is buying a Harley Davidson as your first motorcycle since you were 16 at age 49 a midlife crisis issue?
What about falling back on good ol' human minds? When someone rates a movie the site should ask them why they rated it this way to narrow down certain characteristics. For example, if it is an action film and the user rates it poorly the questions could be:
"Was there too much violence in this film?" Y/N
"Did the plot seem too far fetched?" Y/N
"Was the overall production quality of this film (acting, special effects, editing) not to your liking?" Y/N
"Is this genre of film one you don't normally care for?" Y/N
Etc...
As long as the questions are yes/no and there are 10 questions or less, I'm sure users would be willing to answer them most of the time if they were aware that it was going to enhance the quality of movie selection for them.
Support Liberty, Support Ron Paul
How do you make Netflix better? Well, I think the answer is obvious... replace Netflix with internet-based distribution where a user can download movies to their computer and then stream them to their TV.
Hey, I'm a freakin genius! I'm going to be rich!
*reads Apple Expo Paris press release*
Oh, snap.
What is the url at netflix with the information about this database and access and submissions and rules, etc?
"But saying all romantic comedies are alike because they are romantic comedies is wrongheaded."
It all depends on how dedicated to the genre you are. If you liked everything from Event Horizon, to Evil Dead, to Bram Stoker's Dracula, then there probably aren't too many horror movies wouldn't at least mildly enjoy.
But if you only ever rent Action movies if they star Tom Cruise, then your high War of the Worlds rating wouldn't necessarily imply you'd also enjoy Armageddon.
$8.95/mo web hosting
From the rules at http://www.netflixprize.com/rules : "Residents of the province of Quebec in Canada are ineligible to participate." They go on to list the remaining Axis of Evil and some other countries the U.S. doesn't much care for.
But what's wrong with Quebec? I would presume that they passed some sort of IP law that would make it problematic for NetFlix if the winner were based there.
Exactly the problem with Pandora's recommendations.
The people in the theater would be buying the DVD, all of them, just bump up the ticket price a dollar or two, which is more than enough to cover the disk once they stamp them out by the millions. It's also a loss leader type deal to get them back into the theaters, all of the theaters have been complaining of less and less crowds, a disk with the ticket is a nice inducement, a very nice one. If the real counterfeiters-the pirates-can make dupes and profit at two or three dollars, so can the movie producers. They just don't want to..they want the same cash they used to get for more complex and expensive to produce VHS tapes-they haven't dropped prices in a long time now. I'll keep saying it-volume sales work and we have the technology now to make that happen. And people will have less and less incentive to pirate and file share when they can get legit copies on stamped disks for cheap from a variety of venues. And then they wouldn't have to try and keep coming up with weirder and weirder DRM schemes, again, pissing of their customers and just making things stupider than what they have to be. Humans just don't like being gouged or taken advantage of, that is half the reason no one cares about any ethics when it comes to file sharing, they see it as the producers charge such a rip off price and have been so recalcitrant about it that they lost any ethical business high ground. They have been operating as a blatant price fixing cartel for a long time now, they are angrily getting inj the face of every customer out there with their bogus warnings in front of the movie, and if they just stopped that and dropped prices to a much more reasonable level, and introduce some incentives that make owning disks conveninet for the consumers-like getting to keep the disk instead of turning it in with a rental, the free or near free ones at the movies, etc, they would sell more disks, a lot more,and I bet they would make more net profit in the long run. That's just an opinion, but I bet it would happen.
Deja vu Netflix!
The First one is the age olde "frame problem". This is IT taking a perfectly good database and expecting it to be an even better recommendation system.
Airlines hit the same wall decades ago. They had databases of flights, seats and routes - all excellent. But they really wanted a reservation system based upon ticketing against that database. They finally recognized that nothing less than a mainframe was needed.
The answers you get is all in how you frame the question. Starting with "database and PC" is not going to get to recommendation system without abandoning inherent limits within IT's reference frame.
The second one is CCC Trucking Co solution. Sam owned Crete Carrier Co trucking. Sam owned a significant segment of his market ~50% but wanted to know *how* to grow to 80% of the market for his company. Sam's soluton was simple... he grew 1% at a time, buying smaller truckline competitors. 1% here, 3% there and soon CCC had real marketshare.
Netflix need only implement 1 and 2% solutions. Pretty soon they have a real 10% solution.
The problem with recommendations is that it can only determine what to recommend based on what you've rented, what you've marked you like and what other people who rented the same stuff liked also.
They are leaving out a whole aspect of psychology. The problem is that they have a 2 day lead time to get the content to you. So, it's not just a matter of them asking what your mood is and then providing you with the movie. Instead, they have to predict how you will be feeling 2 days from now and send you the movies.
They should start a voluntary rating program, wherein you sign up and they send you random movies, which you watch and rate (like a critic). Based on your responses to stuff you NEVER would have seen, they have a whole new set of data.
Then there's other stuff, like location. Where in the US are you? Maybe New Yorkers are more likely to like an art movie while a movie about the South is more liked there. What about the time of year? Obviously during holidays you have an easy selection. But certain movies rent more in the Spring and Summer, etc. If you say you're a college student then it knows that you want college flicks in the fall.
Weather? The system looks at the weather forcast for regions (doesn't have to be complex, just a statement like "Region A" is going to be rainy next week") and then queue up more videos with rain in them (or sun, if they are a person who rents the opposite of the weather).
See, you have to play off the cues that make people actually want to watch a movie, not about what they already watched. That's the totally WRONG way. If you just got done watching Ep1-6 of star wars, why the hell would you want another space movie?! After 12 hours of space movie, you're probably looking for a good comedy. In this case, the past has no bearing on the future.
This system CAN work for a place like Amazon. It works well for non-fiction. Say I have a land project and I order some developer books, it can reasonably assume I'll want the more advanced reference eventually.
Would you watch it again? It should ask you about movies you've returned if you're likely to watch it again in the next month, quarter, year, etc. Then it can determine if you want to watch something like it in the near future.
I agree with the TV show thing, keeping episodes in order is a good idea.
Anyway, it's going to need to go out and get other information besides just what people watch. Picking a movie has so much more to it.
Cool! Amazing Toys.
Press release: http://www.netflix.com/MediaCenter?id=5368
Offcial registration and competition information for the Prize: http://www.netflixprize.com/
From the Rules:
To prevent certain inferences being drawn about the Netflix customer base, some of the rating data for some customers in the training and qualifying sets have been deliberately perturbed in one or more of the following ways: deleting ratings; inserting alternative ratings and dates; and modifying rating dates. However, the Cinematch RMSE measured on the final, perturbed dataset does not differ significantly from the RMSE measured on the unperturbed dataset for the purposes of Grand or Progress Prize qualification described below. The RMSE values reported below represent the RMSE measured on both the perturbed and unperturbed datasets to the precision specified above.
Netflix assumes that because Cinematch's RMSE on the perturbed data does not differ significantly from its RMSE on the unperturbed data, the same will be true for the algorithms entered in the contest. This is probably valid if the contestant's algorithms are similar to their own. But they may not be similar. It is possible that Cinematch is a near-optimal refinement of a certain approach to the problem, and only a radically different approach will be able to improve significantly on its performance. But that approach, unlike Cinematch's, may have its performance significantly degraded by the perturbation of the data. Thus the perturbation of the data may prevent a win.
From the rules:
Contest begins October 2, 2006 and continues through at least October 2, 2011.
your submitted predictions that year must be less than or equal to the accuracy value established by the judges the preceding year.
Okay, so the contest doesn't have an ending date, and the judges can modify the winning criteria as they see fit. I hope whoever spends time on this enjoys having others profit financially from their free labor.
Dan East
Better known as 318230.
This seemed like a fun little project, so I registered on netflix and downloaded the file (which decompresses to almost 4gb worth of numbers.) They give you a list of movie IDs, user IDs, and ratings. Your job, if you choose to accept it, is to guess the correct ratings (1-5) for a random slice off a server log after the ratings have been cut out. If you can guess the missing ratings for various movies better than the system can.. you collect a cool $1 mil. The problem is that 4gb worth of user ratings is worthless without more data. You can average the ratings or match users up with others that rated movies similarly.. but that's about it. Having users fill out a small questionaire, pairing them up with others with similar tastes, and using their purchase histories would be infinitely more useful and less complicated than doing it with 4GB worth of numbers 1 through 5 alone. I really hope they use more data than this for their current method of recommendations. (and if they do, shouldn't they release that data if they expect people to come up with a better algorithm.. privacy concerns aside?)
Yup, and there's the related problem of outliers: when you order something out of your ordinary habits. Example, I ordered a bunch of baby books as a gift for some friends on amazon.com. And although I have zilch interest in the little critters, Amazon now keeps recommanding whole ranges of baby stuff.
Non-Linux Penguins ?
So use connotative association.
;-)
The current recommendation systems (if I understand them) are trying to suss out info by comparing the differences between the preferences of people. The problem is that each movie is a very complex mix of themes and connotations.
You need more information on the objects you are recommending to each person. Rather than just compare the preferences you want more information about the object than a rating of 1 to 5. You can do this with genre's, to a degree, but even genre's are imprecise. Is it SciFi or Romance? Is it more scifi than romance?
Think of an object that has width and breadth in a multitude of different dimensions. A Venn diagram meets a hypercube. But where do you get that information?
Why the same place/source that artists and designers do, DUDE!
Oddly enough, to me, the answer comes from a design class: Connotative association. When we are working on a design for a project we start sifting through media. Songs, books, images, ideas, anything that give us a "hit", Makes us think of that project. Think of it as brainstorming but you are sifting through contemporary cultural artifacts and your own psyche to "find" a commonality. Once you have all the media you start to sort it. Does this image work with this sound, does this clothing choice work with this color. As you winnow down those choices you start to get coherent choice of color, mood, and theme that you may not have known existed but thanks to your subconscious you have found a workable choices that should relate to a contemporary audience.
If artistic design uses that method to make new harmonious or understandable art to a culture as a whole than hooking into that same process will allow for better understanding, definition and recommendations.
So you construct a series of images, sounds, media and you use that to classify movies. Popular songs to classic folk hits. Artwork masterpieces to new artwork. All forms of media and periods. Hitchcock is going to "hit" with a black and white pinstriped suit and art deco design elements as will "Hudsucker Proxy" which are both movies that I would enjoy. If you delve into my connotative associations you will find others that are similar and start to generate a mathematical "map" of what I might enjoy based on those associations. You may find that different genre's and movies have similar associations that make them worth recommending. You can also pull from the mappings of other folks connotative associations to see how what they've seen might be of interest to me.
All in all you are just defining a new space. The idea, for me, is easy but I'm not a mathemetician or a database designer.
The downside is getting folks to go through a series of "warshack-esque" media to classify a film. I don't even like filling out the number of stars let alone listening/viewing/reading 5 media bits to relate them to a movie I've seen. But I think it's a viable way to go. The current recommendation system can approximate the above after comparing thousands of films/preferences but without relating it to other media and moods it won't be able to achieve more resolution or granularity.
In the end all art is merely a conversation of a time between the folks who lived in that time based on what has gone before. You are simply trying to better mathematically understand the deeper underpinnings of that conversation to make recommendations. Easy... Right?
"Don't fear death... fear not living..." -me
We used to call this MovieCritic. It used a system called LikeMinds to basically mine their database of users' movie preferences in order to define relationships between the tastes of various users. The theory being that if you feel pretty much the same way about a lot of movies as particular other users do, you're pretty likely to enjoy the movies they liked but which you haven't yet seen. I never had it recommend a movie I didn't enjoy, and I was pretty surprised by some of its recommendations. I saw movies based on MovieCritic's suggestions that I never would have bothered with on my own, and really enjoyed them.
MovieCritic shut down in 2002, and the LikeMinds technology was apparently acquired by Adobe, who as far as I can tell are using it for toilet paper.