The platform combines the first 30 to 60 minutes of Facebook and Twitter data with benchmarks such as website visits and referrals to predict the number of pageviews a story will receive by the end of its “effective lifetime,” which averages around three days, according to Zawya. The model consistently improves by learning from the results of archived stories.
“One of the main conclusions from our research is that social media reactions cannot be ignored when producing traffic predictions,” said Carlos Castillo, senior scientist in QCRI’s social computing team. “You need to take into account not only the number of Facebook shares and tweets each article receives, but also the richness of the discussion around an article in Twitter. This leads to much more accurate predictions than simply extrapolating from current page views.”
The platform will presumably be used in behind-the-scenes decisions about story placement and promotion. But relying on such a model may be dangerous, said Kenneth Cukier, data editor of The Economist and co-author of Big Data: A Revolution That Will Transform How We Live, Work and Think.
“I think it’s risky, to be honest,” Cukier said in a phone interview. “It’s risky because there’s a big presumption in it, and that is that the slow-burn story is not going to catch fire.”
Cukier added that many books don't make best-seller lists until well after their launches, and there’s potential for journalistic articles to exhibit the same phenomenon. Editors should review the platform’s predictions, but they must also be able to defy them and devote space to stories that are important to their publications' identities, even if algorithms don't predict they’ll peak.
“If they’re accurate with 85 percent accuracy, then 1 out of 10 articles are going to bump the trend,” Cukier said, “and they’re denying them to readers prematurely.”
Cukier added that as reader habits change, FAST’s model must too or else it will become ineffective.