Article tiré de la revue The Economist.
EXUBERANCE about both “big data” and Twitter has of late given way to a degree of scepticism among commentators. Yet Dataminr, a New York startup that analyses the 500m or so tweets sent out daily, goes from strength to strength. Founded in 2009 to scour the Twittersphere for important events and news not yet reported by the mainstream media, the firm now has dozens of customers in finance, the news business and the public sector. In January it and Twitter struck a deal to provide alerts to CNN. In April its tracking of tweets was part of a strategy by the authorities in Boston to avoid a repeat of last year’s terrorist attack at the city’s annual marathon.
Dataminr scores every tweet using about 30 indicators of significance, says Ted Bailey, who launched the firm with two former Yale roommates. Dataminr started out seeking market-sensitive information for investors. It now supplies around 50 financial firms. Then came various arms of government; Mr Bailey claims ten clients, though he will name only the city of Boston.
Sorting out the signal from the Twitter noise is not easy. How to tell which tweeters are on to something, and which are just twits? Is there a clue in their language, or their location? Can a false positive be spotted fast enough?
Much of Dataminr’s early work involved tracing back stories that proved genuine, and market-moving events, to find their earliest appearances on Twitter, to help predict which tweets and trends are likely to go viral. For example, it spotted negative words being used in tweets about Apple shortly before a downturn in stockmarket sentiment about the tech firm, says a hedge-funder who uses Dataminr alerts to inform his trading.
Sometimes as few as three tweets send a strong enough signal for Dataminr to alert customers. It reckons its typical edge is around five to ten minutes. It sent an alert about the death of Osama bin Laden 23 minutes before any news network reported it, based on a credible-sounding tweet saying he was dead, and others saying that Barack Obama was calling an unscheduled news conference.
Dataminr sends hundreds of alerts a day, but only a fraction of them are sent to each client and even fewer are truly urgent. At first, Dataminr needed “a lot of help understanding how we source things and what we consider significant breaking news,” says Kenneth Estenson, the head of CNN’s digital operations. But now its alerts are “quickly becoming a core part of our newsgathering apparatus.”
Dataminr has proved it can find valuable information in the Twittersphere. The question is, will it be valuable enough to reward the $50m of venture capital raised so far, let alone the further funding likely to be needed before it is ready to go public or be sold? Mr Estenson says that of the few dozen Dataminr alerts CNN gets a day, “not typically but sometimes it is the first we hear of something.” The hedge-funder sounds more positive, saying that maybe a couple of times a month he gets a Dataminr alert worth acting on, and that it only takes one good trade to pay for the service.
Dataminr is one of a growing number of firms built on analysing data from Twitter, though most do not have its focus on real-time news alerts. “Dataminr’s technology is very advanced; every day there is another example of how far ahead they are,” says Vivian Schiller, a Twitter executive.
Plugged into the fire hose
Other leading firms in tweet analytics include DataSift and Gnip. Like Dataminr, they are among just a handful of firms with full access to the “fire hose”, Twitter’s live feed of its entire output. Unlike Dataminr, DataSift and Gnip are wholesalers of their data to hundreds of other analytics firms that typically use the information, combined with data from other social media, to tell business customers whether and how they are being talked about online. Among social media, Dataminr focuses exclusively on Twitter, it says, because that is where news breaks.
In April Twitter bought Gnip, which may spell trouble for DataSift. But Twitter and Dataminr say this has no implications for their relationship, despite some pundits speculating that Twitter is changing its strategy for working with outsiders. Dataminr’s access to the fire hose is “rock solid”, argues Mr Bailey. He now wants to supply other industries for which real-time alerts could prove valuable, from health care and advertising to risk-management. Twitter itself hopes that Dataminr, by showing the public that Twitter can be a source of valuable news, will encourage more people to tweet about what they see going on around them.