London, December 3 : Researchers at HP Labs in Palo Alto, California, have come up with a way whereby website owners can predict which news stories, blogs or video clips will prove popular in the long term, something that can allow them to allocate extra bandwidth if they need to.
Bernardo Huberman and Gabor Szabo highlight the fact that the number of hits an online item receives when first published is often relied upon to predict its future popularity, but such forecasts tend to be inaccurate as daily and weekly fluctuations in overall website traffic can skew the results.
They suggests that websites instead focus on the rate at which an item picks up views when first put online-suitably adjusted so that views when traffic to a site is low are given more significance than when it is busy.
The researchers say that they have already used this approach to predict that subsequent popularity of 90 per cent of the content on the video-sharing site YouTube.com and the news aggregator Digg.com.
The team was conscious of the fact that Digg stories generally peak in popularity after a few days, and YouTube videos can be much watched for months. They, therefore, trained their software on detailed data from each website to determine the exact relationship between initial activity and long-term popularity.
According to them, the application of this approach to views of Digg stories within two hours of publication enabled them to predict the popularity of those stories 30 days after their appearance to within 10 per cent of the actual number of hits.
The researchers say that they achieved comparable results for YouTube clips too, reports New Scientist magazine.
HP's approach isn't the only method of forecasting traffic, as video search engine blinkx.com also claims that it can make similar predictions by analysing the number of distinct sites to which clips are uploaded rather than the number of initial downloads.
Their approach allows them to identify the most consistently popular videos, and put them at the top of their search results.
Alan Mislove from the Max Planck Institute for Software Systems in Saarbrucken, Germany, considers this work important because it may point web advertisers to the most viewed areas of a website, and be useful in deciding how much bandwidth to assign to individual videos.
"If you knew a video would only be popular in Europe, you could decide not to host it in the US," he says.