What movie should you watch tonight? Personalized recommendation engines help millions of people narrow the universe of potential films to fit their unique tastes. These services depend on a machine-learning strategy called singular value decomposition, which breaks down movies into long lists of attributes and matches these elementsto a viewer's preferences. The technique can be extended to just about any recommendation system, from Internet search engines to dating sites.
For more on machine learning, read "How To Teach Computers To Learn On Their Own" in the July 2012 issue of Scientific American.
Illustration by Jen Christiansen; Interactive by Ryan Reid
See what we're tweeting about
10 Comments
Add CommentNo, movie recommendation systems do not work, they do not work at all; well, they do one thing very well: they make hysterical lists of non-related films, for your amusement that have NOTHING to do with taste or movie evaluation. I believe my understanding and appreciation for movies is the result of a lifetime of viewing both for pure entertainment, and as an art form. There are criteria for making "good" movies that have never, not once been able to be put into ANY recommendation list.
Reply | Report Abuse | Link to thisThe "lists" are absolutely useless; I use netflix, and watch almost exclusively films that would be considered low box-office draws, independent films, creative films, well made, well directed, well acted, well photographed, well lit, etc. Also good stories with good writing, believable acting, and innovative and unusual cinematic approaches. So here's my "top 10 list" from them:
1. Adventures in Lalaloopey Land
2. Dr. Who, which I already have in my queue...
3. The Veteran, which I already started and then rated not liked
4.The Super Hero Squad Show (did I mention I rate ALL cartoons as 'not liked"?)
5.Ralphie May - hate him (sorry Ralph.)
6.The Heir Apparent - OK, this I would watch
7.The Veteran AGAIN
8. Ralphie May AGAIN
9. Turtle's Tail (remember I vote all cartoons down)
10.Bones Which I really really hated - totally unbelievable characters, plot lines that blow up in their own faces, leading characters with personality flaws requiring immediate and intensive therapy that are of zero interest to me, etc., etc.
No, recommend lists are jokes, no help, and only serve to urge me to fling the remote out the window.
If they began rating movies on REAL human criteria, not "Like JAWS", there might be a point to them, but using dumb (really dumb) categories, like "Quirky, foreign Comedy" does no good at all.
Companies that try to generate lists display a profound misunderstanding of how the human mind works - using their logic, if my car is red I must only like red cars...AND I would want to DRIVE ONLY RED cars, and I would want RED CARS recommended to me, because, after all, I looked at one once... totally useless endeavor, "recommending" films through computer-generated lists.
Is this a science article, or a self-congratulatory display of an attempt at interactivity?
Reply | Report Abuse | Link to thisIn attempting to disaggregate movies into their component parts and then match based on these parts, this type of move recommendation engine loses something in translation. Agree with promytius that there is a richness in film that is inherently subjective and requires human interpretation.
Reply | Report Abuse | Link to thisAn alternate and more powerful recommendation engine would be based on collaborative filtering, a technology spun out of the University of Minnesota in the late 90s (into a company that served primarily dotcom retailers and today no longer exists - Net Perceptions).
Collaborative filtering works by finding other people who like virtually the same movies you do. These "taste soulmates," known only in aggregate and anonymously through their rating history, preserve the subjective richness of personal taste and turn out to be a much better proxy for movie preferences.
If anyone wants to try this, take a trip to www.movielens.org, which I believe is still run by UofM. As you might expect with a non-profit university-run site, the interface won't be much. But seed the engine with a few of your favorite movies, and see if the recommendations don't begin to hit the nail on the head for you.
Nice link, better than Netflick's.
Reply | Report Abuse | Link to thisA long list of attributes is not enough to power truly meaningful discovery. The engine must be able to truly understand what each the mood, plot, etc. of each title and there relative dominance. Ideally the engine should also have a similar understanding of the viewer's taste.
Reply | Report Abuse | Link to thisThe problem with Collaborative Filter technology is that it is absolutely blind to the very human meaning that is required. It sums and averages all users and their tastes, but their is no such thing as 'average' taste.
The Jinni Entertainment Genome uses Natural Language Processing and machine learning to automatically extract the meaning of each title in the form of 30-50 highly descriptive 'gene' tags.
Take a look at the genome (right side) for 'Alien' to see an example of how richly a title can be describes beyond 'Sci-Fi) http://www.jinni.com/movies/alien/
The article left out the "how" - i.e. the process skipped in step two.
Reply | Report Abuse | Link to thisNot that it matters, I generally find that the greater statistical basis a recommendation engine has, the more useless the recommendations. This is hardly surprising.
You are right! People's taste in movies (or anything for that matter) is a little more complex that statistics. All these algorithm based methods are just blind guessing. There is no such thing as 'average taste'.
Reply | Report Abuse | Link to thisTo recommend content the engine must understand meaning the ways humans do.
http://www.jinni.com
This is a very much simplified representation of how such systems work. The fact is, they do work sufficiently well that businesses profit from using and developing them. That they don't work for you may be relevant; perhaps another system would work better. However, they work well enough in most cases that they continue to be used. You may not like the fact that your actions can be reduced to statistics, and that those statistics can be used and are used to predict some of your actions, but that is the case. Being in denial about it won't stop that being so, I'm afraid.
Reply | Report Abuse | Link to thisRecommendation engines become more accurate as they get more data points on the movies in question, which they can tie to you, the individual user - if you like mostly low-run, lesser-known movies, then yes, tragically the standard recommendation engines probably do not have a enough data points to find those rare 'unwatched' movies that match your favorite 'watched' rare movies. But, they do work in general . . . search auto-suggestions work in a similar way, albeit with less personal data points for you, the individual user ... but, do you find them helpful for everyday search terms? (start typing terms in Google) Do you also notice that if you search for more rare topics, the auto-suggestions get less accurate? If so, movie recommendation engines work similarly, but, probably just have less data on the movies you like.
Reply | Report Abuse | Link to thisSince moving to the Philippines I found very little English language programs on free-to-air television, so ended up getting cable. I started to realise how much Hollywood type-casts actors for certain types of movies, so now I look to see who is in a movie to get a fair idea if it is worth watching. I get bored easily, so want mainly action. Anything with Denzel Washington or Jason Statham is good. But I saw one listed that had Robert De Niro and Bill Murray. That sounded like strange chemistry, so I thought I would check it out. It was odd, and I ended up switching to something else half way through. I don't know the names of the actors in the Final Destination movies, so may miss some good movies that they might be in. But generally speaking, unknown names mean it is a C-grade movie, and I think my method is the best there is. I found that video stores have insufficient categories to properly describe films. One I went to had the movie Fargo in the comedy section. This movie was based on a true story about a couple of murderers, and the only thing funny about it was the Scandinavian accents the people in that part of America have (like Rose in the Golden Girls had). There are many movies that are listed as comedies, but they don't make me laugh, just the occasional smile. Now if someone can think of a name for all the different categories and publish them, we might start getting somewhere.
Reply | Report Abuse | Link to this