Let's take supercomputer with 1024 Intel Nehalem cores and let's call it Nautilus.
Let's take Nautilus and give it 100 million news articles to read.
Let's ask Nautilus to apply mood detection for those articles searching for words such as "terrible", "horrific" or "nice". Those words will describe global sentiments.
Let's ask Nautilus to detect words for locations, like 'London', and record those locations on the map, showing where the specific waves of news were produced.
What do we get?
We get a trend, showing that BEFORE large events happening around the world, the mood of the words used in the news becomes very negative. And we know the location of the places, where the mood goes down. Such analysis is possible because Nautilus was programmed using similar algorithms for analyzing news as those used in DNA simulations and weather forecasting. It was done a posteriori - after the events happened - but now it will be applied to the ongoing events with better location engine.
What Nautilus did is, indeed, a weather forecasting for the large events in the worlds. Not perfect, but definitely above the level of chance. Still, as @mjrobbins points in his article for Guardian, you need human to make sense out of the analysis outputted by Nautilus, so it's not that computer does all the job. But it's a good start.
Details are in this paper and slightly biased BBC news article.
What do we get?
We get a trend, showing that BEFORE large events happening around the world, the mood of the words used in the news becomes very negative. And we know the location of the places, where the mood goes down. Such analysis is possible because Nautilus was programmed using similar algorithms for analyzing news as those used in DNA simulations and weather forecasting. It was done a posteriori - after the events happened - but now it will be applied to the ongoing events with better location engine.
What Nautilus did is, indeed, a weather forecasting for the large events in the worlds. Not perfect, but definitely above the level of chance. Still, as @mjrobbins points in his article for Guardian, you need human to make sense out of the analysis outputted by Nautilus, so it's not that computer does all the job. But it's a good start.
Details are in this paper and slightly biased BBC news article.
It's an interesting concept, but I'd refrain, at least for a while, from asserting that it's predicting the future.
ReplyDeleteFrom what I've read in the paper (admittedly I've only read some of it), currently it only predicted past events, which can be misleading in that while developing such systems, the algorithms are usually shaped by the data, i.e. they're tweaked until they predict what actually happened.
My other concern is that if their algorithms are indeed correct, it means that journalists as a whole are able to predict the future.
You are right - Nautilus only processed past news, but they are working on algorithm that draws predictions from ongoing stream of data. I think they have rather cruel results, but there is some pattern, similar to what you see in weather prediction. I especially liked the piece where they narrowed potential whereabouts of Bin Laden (of course still referring to past events).
ReplyDeleteHehe if journalists' focus on specific information is dependent on fulfilling the demand of viewers, then we are all somehow predicting the future ;-)