Slashdot Mirror
Mapping Google News
CousinLarry writes "A neat project called Buzztracker.org has been mining Google News for over a year and keeping track of relationships between geographic locations mentioned in articles.
The results are some really cool maps that actually seem to reflect the "buzz" of the day - check out the Vatican clusters from earlier this month, or the global New Year's chatter. You can also dig down into the articles from which the maps were generated."
It looks like the code needs a bit more tuning. http://www.buzztracker.org/index.html lists Nelson, NZ, as one of the hot spots. Clicking on that lists a bunch of articles about apartheid. I think the site code misinterpreted a reference to Nelson Mandela in one of the articles.
This site has another list, of the sources Google News uses (something Google refuse to publish). Also an interesting use of data mining.
Give a man fire, and you warm him for the night. Set a man on fire, and you warm him for the rest of his life.
I actually spent the last 3 months of my life writing something very similar to this for my PhD thesis. My work was slightly different then what you explained though. Basically I take the Google data, use it to prime the network, and then feed in historical data from a particular news site into the neural network. The app processes the data, and predicts which news events the news site will report on in the coming days. I've run this application against Slashdot, since such a wide range of topics are reported on here, and have found that the application can guess 7 stories from the next day 87% of the time. I didn't have nearly this much success with other news sites, so I decided to figure out why I was so successful. I found that the nueral network was simply reporting on news events that happened more than 3 weeks ago, contained words like 'Star Wars', would search for anything about Google and then would add the question "Are they becoming evil?", would take all Microsoft and EFF press releases, and somehow managed to pull every 17th email from Linus Torvald's inbox, would repost every 19th article, and would occasionally take a story about someone being prosecuted and insert "Your right online" in front of the original news source's title. Unfortunately for me, the nueral network seemed to learn too much from Google and now requires that I become a member of its club before I can see any more future stories.
What we have here is one computer algorithm aggregating another computer algorithm's assessment of "newsworthy," with no provision for hindsight or fluff-vs-historical weighting. It's a neat idea, and the graphics are pretty slick, but I don't see any real value here.