Twitter as Ghetto Prediction Engine

Twitter predicts future box office

from kottke.org by Jason Kottke

11 people liked this

A study by researchers from HP's Social Computing Lab shows that Twitter does very well in predicting the box office revenue for movies.

[Researchers] found that using only the rate at which movies are mentioned could successfully predict future revenues. But when the sentiment of the tweet was factored in (how favorable it was toward the new movie), the prediction was even more exact.

But as someone noted in the comments:

Works fine until people realize it works, then they start gaming it, and it stops working.

Prediction Conviction

First do yourself a favor and read this article: The Predictioneer

For the lazy, basically this fellow Bruce Bueno, a professor at New York University, has been developing a model to predict such seemingly unforeseeable things as who will win an election, or if a certain bill will make it through congress, etc. While avid political scientists might be able to make similar claims, Bueno has the statistics to back it up: over 90% accuracy at this point, with an even better model in the works.

So how does he do it?

1. Amazingly informed starting information which he gathers by paying attention to what's going on in the world of politics [for example: "Give a shit factor for health care" Obama 98/100, McCain -77/100, ... "Able to do shit about it factor" Obama 60/100, McCain 5/100, Clinton (Bill) 80/100 ... etc.]

2. Game theory running software, which runs all known variables against each other and predicts an outcome.

Really, it's that simple. The difficult part is obtaining the correct input data, because the engine is only so accurate as what you provide it, which is why Bueno, being an expert in the world of politics, is able to achieve such high results.

Now.

Take something really similar to an election, say high school prom. We want to predict who will end up taking who. Even without knowing English. Even without being able to see the kids to analyze how objectively attractive they each are. Even without knowing who's friends with who on facebook, etc. If we could just track relative movement over time for a number of weeks, I bet that would be enough data to run an accurate model. Luckily Japan has given us this.

So now take a similar example: a night at a club. If we could track movement, we could probably come up with a 10% accurate model of who would go home with who. If we could couple this with eye tracking, we could probably bump it up to 50%. If we could add past relationships for the people involved, to see what facial/personality features they prefer in a mate, we could probably tip the scale closer toward 90%. Now switch into first person- if you're in the club and you want to know who you might have a shot with, then you can work with the software by inputting good data just like Bueno [example: Suzy likes Bill 75/100, Suzy finds guys like Bill attractive ?/100 (now you show the system her past boyfriends on facebook and it analyzes their physical appearance for similarity = 99/100) ... etc. Run this on everyone present, and you can figure out who to talk to. Just keep in mind that everyone else is probably doing the same thing.]

Take something less trivial like who's a compatible lifelong mate, and we'll be able to cause all sorts of mischief.

Cyborg Socialization

Statistical Mastery

Cellphone traces reveal you're so predictable

We may all like to consider ourselves free spirits. But a study of the traces left by 50,000 cellphone users over three months has conclusively proved that the truth is otherwise.

"We are all in one way or another boring," says Albert-László Barabási at the Center for Complex Network Research at Northeastern University in Boston, who co-wrote the study. "Spontaneous individuals are largely absent from the population."

Barabási and colleagues used three months' worth of data from a cellphone network to track the cellphone towers each person's phone connected to each hour of the day, revealing their approximate location. They conclude that regardless of whether a person typically remains close to home or roams far and wide, their movements are theoretically predictable as much as 93 per cent of the time.

--
Just another instance of the machine understanding us better than we understand ourselves. Like renting a movie even though netflix says you won't like it. A few days later the formula laughs as you give it a low rating. The god in the machine knows you like no one else.

"Indeed the very hairs of your head are all numbered. Fear not therefore: ye are of more value than many sparrows."

So much more valuable in the whole political economy [politonomy or ecolotics?] of things. $o much more valuable...

Just a Word About Food

2008

Every restaurant should have a complete list of the dishes they serve available online. This should be hooked into an unaffiliated food profiling service which keeps track of what you eat and how much you enjoyed it like netflix. As this starts to build it would become quite useful. Say you walk into a restaurant that you've never been to before. Instead of searching through the menu in order to guess at what sounds good before the waiter starts to get impatient you could simply consult your profile via your cellphone. You tell it the name of the restaurant that you're at and it gives you a list of the top ten dishes it thinks you'll enjoy based on what you've rated in the past and what "connoisseurs like you" have said about this restaurant's various options. On a more practical level it also keeps track of allergies and disdain for particular items: "Thank god it told me that they make the sauce with soy milk, because I'm so allergic that I probably would have died." It would also tell you things about yourself that you wouldn't have figured out otherwise: "Based on your hatred of these ten dishes, it seems you do not like their shared ingredient, cilantro." Or say you get sick after you eat certain foods. You let it know each time this happens and it examines common ingredients to figure out that you're allergic to eggplant. Next time you ask them to leave it out. Problem solved. You could also see trends of food poisoning which would tell you where to avoid. It seems that it would be impossible to keep neighboring restaurants from sabotaging each other with fake claims, but maybe we'll find a solution- humanistic input and participation requirements would protect against primitive bots which might make large scale sabotage too inefficient to become a problem.....

Update:

MenuPages Brings Restaurant Menus to Your iPhone [Downloads]

from Lifehacker by Lisa Hoover

18 people liked this

Pandoras Disk Jocks

3.09

Short: Since we all have musical profiles, there should be a way to listen to universally appealing music automatically, like a lowest common denominator for every possible audience. Extending this, when we go to clubs we should upload our profiles so that the dj would mix more efficiently and the music would change with the type of people in the room.

Long:

Many of us have an ever growing musical profile attached to us via Pandora, Itunes or similar such programs. These keep track of what songs we listen to and how often. In the case of Pandora, the service actually tries to understand the musical "taste" of its users based on hundreds of criteria (repeating form, sad lyrics, solo guitar, etc.) Since most of us have these profiles, we should use them for more than our own interest i.e. "did I really listen to that song six times yesterday?"

Pandora already suggests music that it thinks you would enjoy (similar to Netflix), but much more could be done. It's amazing we aren't doing this already. For example: I'm with a friend. We hop into a car to go on a road trip. Obviously music is going to be an awkward problem on a long trip with someone who's tastes might be completely different. As a solution, instead of us taking uneven and annoying turns trying to guess what would be most palatable for each other, we simply log into BOTH of our Pandora or Itunes accounts simultaneously. These two profiles then compare notes on our listening habits and create a playlist based on the highest possible common denominator: what will make both of us most happy and the least annoyed. Though we each have unique collections, there are probably a dozen or so albums that we both happen to listen own or listen to, which would be more than enough music to last a days worth of driving. It could also determine characteristics that we both appreciate in music and try to add variety from there (like the music genome project). Obviously it won't be perfect but this will be remedied through the manual skip of bad songs (like a veto by either party). The service could also just provide a list of the most congruous thousand or so songs which we could look through and select from manually.

Expanding this model the same process would work with three, four or more people. For house parties instead of someone manually making all the decisions, everyone would log into their accounts when they arrive, thus changing the mix as different crowds come and go. This could even work in a club. Hundreds of people could contribute their musical desires automatically creating a scene that would change as often as the people within it.

"...man's relationship to his environment has changed. As a result of cybernetic efficiency, he finds himself becoming more and more predominantly a Controller and less an Effecter"

-Roy Ascott, 1964

The DJ then becomes an interpreter of this massive amount of information. He might see that many people present have recently started listening to a certain popular song. Because he knows that they'll enjoy dancing to their new found titillation, he mixes the song's chorus a number of times throughout the night, combining it with other songs thus creating a remediated version of something he can be sure people like. His melodic insertions also become an interesting form of communication. Suppose someone with uncommon preferences is in the building- we'll call him Fred. Our DJ sees that Fred mostly listens to obscure jazz from the 60's and 70's and that he has listened to Archie Shepp's Attica Blues album nearly 10 times in the past week. Our DJ also likes this obscure subsection of the musical world, which is why he noticed Fred in the first place (this collection of profiles shows him the most congruous/unique individuals in the crowd). Because Fred had obviously taken a recent liking to this record, our DJ takes quick listen. It's good. The first track is a type of funk tune that features a great drum intro. He quickly cuts this intro, loops it and mixes it in as a background to an acapella recording of a pop track. Of course Fred would be delighted and, as a way of saying "thanks for turning me on to some good music", our DJ even sends Fred the recorded mix with his loop in it as a type of memento and personal leitmotive. The next time Fred comes out and logs in, his profile will be tagged with this recording that the next DJ will be free to include or exclude at his discretion.

More than a simplistic improvement in song selection, this tool would show what people had listened to, how many times, and how recently. Therefore, to the trained interpreter it would reveal the actual mental state of the individuals and the crowd itself. He actually sees what is fresh in people's minds and therefore ripe for manipulation and artistic communication. This all plays into a mental commodification of attention and relevancy. Things that are more pervasive in people's recent memory carry more sway and could theoretically affect on an unprecedented scale. A club's atmosphere would also change in real time with the feeling of its patrons.