| |
|
I'm not really sure of the best way to begin or invite discussion about this, so I suppose I'll just describe it the best I can and hope others have comments or perhaps have heard of this already.
There's a site called 'Half Past Human' whose creator is up to something which is in my opinion quite interesting. He harvests massive amounts of textual data sampled from real communications on the Internet using web bots - Usenet postings, forums, news sites, blogs, and just about anything that doesn't use a "robots.txt" to tell them not to. He then apply sets of lingual rules and filters to "distill" the information. An example of the conceptual underpinnings of the practice can be found here.
He uses sets of interpreting rules to generate predicative information out of the mined data. The information is usually somewhat vague, but is generally a no-match for "past events" (in other words specific enough so as not to be universal in any way), and can oftentimes be matched up to a particular event with relative ease post facto. He has at times and in various places been credited with predicting "9/11", the Sumatran quake/tsunami, certain events in Iraq, and other events of varying infamy. He thinks he has at best a 40% success rate when it comes to accurately analyzing the gathered data, and says the engine works especially well with economic and stock-related predictions because these areas react especially potently to the human emotional climate.
To be honest, I can't think of a straightforward and arguable reason why mining the emotional quality of the human experience might yield insight into future events. Still, this fascinates me. Anybody else?
There's a radio interview with the fellow here. Fair warning: It's a couple of hours long. The interview starts around 34:04 in the archived audio. |
|
|
