Real-time web search – which scours only the latest
updates to services like Twitter – is currently generating quite a
buzz because it can provide a glimpse of what people around the
world are thinking or doing at any given moment. Writes Jim Giles
in an article originally published in
New Scientist.
Interest in this kind of search is so great that, according to
recent leaks, Google is
considering buying Twitter.
The latest research from the internet search
giant, though, suggests that real-time results could be even more
powerful – they may reveal the future as well as the present.
Google researchers Hyunyoung Choi and
Hal Varian combined data from
Google
Trends on the popularity of different search terms with models
used by economists to predict trends in areas such as travel and
home sales. The result?
Better forecasts in almost every case.
Projected sales of cars and vehicle parts, for example, can be
extrapolated from the figures for the previous month and this time
last year, but adding in the volume of automobile-related search
queries to the model cut the error rate by 15%.
Early warning
It works because searches reveal something about people's
intentions. For example, Varian suggests a surge in people using
the term "unemployment insurance" may help predict looming economic
problems.
Google has demonstrated before
that search data can predict flu outbreaks, and last week World
Bank economist Erik Feyen
said he could cut errors in a model that
forecasts lending to the private sector by 15% using Google
search data.
But real time results could have even more predictive power:
knowing what people are actually doing, not just thinking, at a
particular instant gives a strong hint of the future
consequences.
Message to the future
Other data is targeted more specifically towards
future dates. At the FutureMe website you can send yourself an email
scheduled to arrive at any date between now and 30 years hence.
Johan Bollen of Los Alamos National Laboratory and
Alberto Pepe of
the University of California, Los Angeles, applied a mood rating
system to the text from over 10,000 FutureMe emails sent in 2006 to
gauge people's hopes, fears and predictions for the future.
They found that emails directed at 2007 to 2012 were
significantly
more depressed in tone than messages aimed at the subsequent
six years. Could they have predicted the world's current economic
slump?
Without more data, that is no more than an intriguing
possibility. So Bollen plans to look at more FutureMe emails, as
well as Twitter messages, to search for mood swings that foreshadow
other economic changes. If he finds any such links, the same
sources might be used to try and predict future economic
fluctuations.
So will our online footsteps become a central part of economic
forecasting? We'll have to wait and see – or perhaps do a quick web
search.