28 September,2010 02:40 PM IST | | Agencies
Web search statistics can be used to predict human behaviour - from record sales to the spread of influenza, found a new study.
But the researchers also found ''traditional'' information sources are just as effective, and in cases more useful, at spotting trends, reports ABC Science.
Sharad Goel and colleagues from Yahoo! Research in the United States wanted to see if web search query logs could be used to predict how well something is going to do in the future.
For the study, they looked at box office movie revenues, video game sales, and Billboard Hot 100 songs over various periods in 2009 to see if Internet search counts could accurately predict what will succeed and what won''t.
The researchers found search-based predictions did provide a good indicator of real outcomes for both movies and video games sales.
But they were only moderately correlated with outcomes for music.
They then compared the data with predictions based on traditional information, such as production budgets, critics' ratings and prequel revenues.
They found these traditional predictors did manage to outperform search-based predictions for movies, music, and sequel video games.
But, search query data did better than traditional methods in predicting the success of non-sequel video game revenues.
Goel said sudden changes in search volume might help to identify "turning points".
He said this type of information might prove useful for applications such as financial analysis, where a minimal performance edge can be valuable.
Matthew Sheppard, research and development manager of Canberra-based IT company Funnelback said the study also demonstrated search queries are better at predicting how music would perform on the charts.
According to Sheppard another area of interest is using web searches to track flu trends.
Sheppard says overall the paper shows normal methods work fine except when there are sudden changes in trends.
The study has been published in this week's Proceedings of the National Academy of Sciences
