Data-mining collective intelligence systems

[bascule]

Okay, I'm sorry if the threads I post are too technical/boring. Seems like most of them go ignored/unanswered, but we're probably in the post-con "No one gives a shit" lull of the forums, I suppose.

Rather than giving a rant here, I'm going to just point out a few web sites. Hopefully, you've heard of all of these:

http://likebetter.com/
http://reddit.com/
http://digg.com/


These sites data-mine collective intelligence. And they all do a damn good job. What emerges on them is the result of a collective decision-making process which is entirely automatic. LikeBetter seems to have constructed an expert system on humans based on collective decision making coupled with a profiling mechanism that lets you report when it has made a mistake, and consequently is continuously evolving. The results it gives are shocking (although for some reason, it seems to think I'm left-handed. Maybe I'm just right-brained)

And think about the kinds of information that can be extracted from the AOL search data which was exposed.

[Asbestos]

Re: Data-mining collective intelligence systems

The likebetter site is amazing. I can't believe how much time I wasted on that damn thing, and how accurate it was. It only got one wrong, and I think that was because I clicked on the wrong image a few 'tests' back.

Anyway, I'd really like to see the results of sophisticated data-mining against the AOL search data. For some of those people's queries, "kill your wife" among other things, it could get extremely interesting.

[theprez98]

Re: Data-mining collective intelligence systems

Likebetter pegged me as a male and nailed my household income so far.

[converge]

Re: Data-mining collective intelligence systems

Definately interesting. I took a little bit of time to check out likebetter.com. It was able to come up with quite a few correct things, however:

It pegged me as a thirty-something that would prefer to speak in front of a large group, that I feel energized by meeting new people. Appearantly I'm also a religious person that regularly applies emotion to my decision making.

I would have suspected it to easily determine that I watch way too many cartoons, enjoy/play music, like pictures with women in them, prefer the color green, etc ... but it didn't. I suspect that a lot of its intuition comes from the yes/no answer you provide to previous guesses, such as its initial guess that I'm a man, then the assumption that I prefer one-to-one contact.

I also found it repeating/inversing many combinations of pictures, either to verify which I preferred, or to force a conclusion based on different choice. It was unable to determine that my pictoral memory functions are up to par.

Then I hit a good hour or so where it came up with absolutely nothing but more picture (although it did not report on my abilities to persist).. I gave up for the night.

[Virosa]

Re: Data-mining collective intelligence systems

The likebetter site is amazing. I can't believe how much time I wasted on that damn thing, and how accurate it was. It only got one wrong, and I think that was because I clicked on the wrong image a few 'tests' back.

Sweet Jesus, that thing thought I was a teenage girl. How many fucking times do I have to click on Voltron to be a thirty something dude?

[robosycho]

Re: Data-mining collective intelligence systems

It could just be that I'm crazy, but it pegged me as a chick who likes talking in front of other people.... And then continued to get them wrong. Maybe I'm just messed up like that.

[Deviant Ollam]

Re: Data-mining collective intelligence systems

i found too many photo options appearing in the likebetter site where i didn't like either one of them at all, or had no clear opinion on the matter. being forced to choose anyway skewed my results i'm certian.

[theprez98]

Re: Data-mining collective intelligence systems

After visiting the site a couple of times I thought that (unless it was way off) you might be more likely to remember the things it gets right and less likely to remember the ones it got wrong. Especially the ones that aren't so clear cut. So that in the end you might think "Wow that site really nailed it!" when in fact it was just a matter of probability.

So far I've found 3 of the 8 were wrong.

[TheCotMan]

Re: Data-mining collective intelligence systems

After visiting the site a couple of times I thought that (unless it was way off) you might be more likely to remember the things it gets right and less likely to remember the ones it got wrong. Especially the ones that aren't so clear cut. So that in the end you might think "Wow that site really nailed it!" when in fact it was just a matter of probability.

Like fortune telling, or Horoscopes?

[theprez98]

Re: Data-mining collective intelligence systems

Like fortune telling, or Horoscopes?

That is a good parallel.

[TheCotMan]

Re: Data-mining collective intelligence systems

Junk food-eating, thirty-something, night person, extroverted, fairly poor, right-Handed, Liberal, Male, who does better one-on-one than addressing large groups, has a few close friends, relies on emotion to make decisions, prefer instructions when they're fuzzy, tends to see the glass half full, with an annual household income is $25-50K, give or take, whose desk is pretty clean right now, chooses boxers when pressed to choose between boxers or briefs, find it easy to empathize with other people, attended college, a religious person, study/work better with music playing in the background, feel drained by meeting lots of new people, the kind of person who acts first and thinks later, probably like to take things as they come, very interested in matters of technology, and am not into books.

Hmm. After a while, no more "pink brains" appear. Been running it for over an hour without no new "pink brains."

Seems that it was only right about 60% of the time, and the future assertions it claimed seemed to be more closely based on previously selected *answers* than images.

Needless to say, the results of this test cause me to doubt the validity of results in this test.

[theprez98]

Re: Data-mining collective intelligence systems

Seems that it was only right about 60% of the time, and the future assertions it claimed seemed to be more closely based on previously selected *answers* than images.

Needless to say, the results of this test cause me to doubt the validity of results in this test.

Pretty close to the results I got. Considering some of these are not so concrete thats not much to say for their accuracy. Still, its an interesting idea.