TMCnet News

Web "betweenness" predicts election results and stock fluctuations
[February 08, 2008]

Web "betweenness" predicts election results and stock fluctuations

(New Scientist Via Thomson Dialog NewsEdge) ACTIVITY on the web can provide more than a snapshot of what people are interested in on a given day. It is also being used, with some success, to predict future stock prices and election results.

Tools such as Google Trends
and Blogpulse
track what people are talking or thinking about by recording the frequency with which words are entered into search engines and appear on blog sites. Now Peter Gloor

at the Massachusetts Institute of Technology is going a step further, and using the web to make specific predictions. His software, called Condor, has predicted the results of an Italian political party's internal election and successfully forecast stock market fluctuations.

Gloor's secret is a property of networks called "betweenness". Condor starts by taking an ordinary search term ? the name of a political candidate or a company ? and plugging it into the Google web search engine. It then takes the URLs of the top 10 hits returned by Google and plugs them back into the Google search field, prefaced with the term "link:". In response, Google returns the sites that link to the 10 original sites. Condor then repeats the process with the new set of sites.

Next Condor maps the links between all the sites it has come up with ? whether or not they mention the original search term ? and works out the shortest way to hop from one to the other via the links they contain. The more often a site is involved in linking different members, the higher its betweenness score. To produce an overall score for the term originally submitted to Google, Condor then averages the betweenness scores for all the sites.

This score alone provides some indication of popularity. In December 2006, Gloor entered a range of film titles from that year into Condor and found that of the 10 with the highest betweenness scores, five won Oscars, four were nominated and one got no award.

Gloor is also working on an improvement that tweaks Condor so that it can be instructed to search only blog sites or chat forums. Once it has a betweenness score for these separate categories, it weights their contributions to get a final rating. He found that the magic combination for the final rating is a 15 per cent contribution from the web, 5 per cent from blogs and 80 per cent from discussion forums.

He suggests that the forums involve people with the most to say: "A forum has all the most interested people talking to each other." Using this version of Condor to track mentions of various companies, he was able to predict three days ahead with up to 80 per cent accuracy whether the companies' stock prices would move up or down.

Francesca Grippa and Pasquale Del Vecchio of the University of Salento in Lecce, Italy, have used Condor to predict the outcome of the election for the secretary of Italy's Democratic Party. A month in advance, they predicted the order of the six candidates nearly as well as the exit polls. They say it might be possible to do the same for the US presidential election.

Despite this apparent success, Roger Jowell, director of the UK's Centre for Research into Elections and Social Trends, has reservations about Condor's ability to predict election results. "In the stock market, people get nervous not because of what they feel but because of what other people feel," he says. "But with elections, people ignore a lot of the press."

Copyright ? 2008 Reed Business Information - UK. All Rights Reserved.

[ Back To's Homepage ]