giovedì 24 ottobre 2013

Content contagion / Internet epidemiology by H.M. @ The Economist 13.Mar.2013



MORE than an hour of homemade video is uploaded to YouTube every second. Although the vast majority of clips are destined to languish largely unwatched on its servers, a tiny fraction go viral. No one knows for sure why some videos spread while others wither. But the first step to knowing why some things spread like wildfire is to work out how this happens. Now a team at Microsoft Research, led by Jake Hofman, has devised a way to measure the virulence of online content.

For 12 months Mr Hofman's group recorded every tweet containing a link to anything on the world's 40 most popular websites for news, music and videos. These featured over 1 billion pieces of content on YouTube, Yahoo, Instagram and the BBC. The researchers then selected those that were linked to on at least 100 different feeds, for a total of nearly 300,000 web pages and 1.4 billion tweets. They then painstakingly reconstructed how each of these stories passed from person to person. This allowed them to identify the different ways information spreads.

By looking at the average distance between all people who posted a link, Mr Hofman and his colleagues assigned each piece of content a score for its virulence, out of 100. Predictably, only a handful of web pages from the original billion got the full 100. But the data sprang some surprises, too.
For one thing, Mr Hofman found that virulence is distinct from popularity. Major news stories might be read widely but would quickly fade from the cultural consciousness. For example, traditional broadcasters might tweet news to a million followers. Often, no more than a few hundred of these will retweet in turn to their followers, a couple of whom might do the same. Soon, though, the story would peter out. Viral content, by contrast, can stem from an obscure feed but, by definition, rapidly gains momentum. Within a few days, it would have spawned many new branches as more and more people share it. Truly pestilent information, about one in a million stories, persists for 20 generations or more.
That may explain why last year's internet sensation, "Gangnam Style", scored less than amateur parodies of Gangnam Style that never reached as nearly many people as the original (which has notched up a record 1.4 billion views on YouTube). Another discovery is that many people share tweets without ever following the links they contain. While an online Twitter campaign to end malaria went massively viral, the video that inspired it racked up just a few tens of thousands of views.

On March 5th Microsoft Research demonstrated an application called ViralSearch that visualises the virulence of content in two complementary ways. The first depicts the family tree of who retweeted what. In the second, each piece of online content is represented by a circle. The size of the circle represents how many people it reached through Twitter. Concentric shading reflects the number of generations it spawned. Users can then drill down to compare the popularity and virulence of different tweets linking to the same content, say, or focus on various tweets posted by a single user.

The technology could eventually find its way into Microsoft's Bing search engine, allowing users to hunt for the most viral online content rather than merely the most popular. Conversely, anyone fed up with the Harlem Shake could opt to exclude such material from their results. Even the prototype would prove be a boon to musicians, writers and marketers, who could comb Mr Hofman's data in an effort to work out what precisely determined the extent of online contagion. Unfortunately, there are no plans to make it publicly available.

Full-blown viral search will take time to develop. The billion or so tweets Mr Hofman looked at over one year are now being generated every couple of days. Nevertheless, Microsoft is currently scaling up its Twitter monitoring to analyse billions of stories, and billions of relations between Twitter users, with the aim of ultimately accommodating every website on the internet in real time. Until it does, discovering new viral content it best done the old-fashioned way: just ask a teenager. Read more

Nessun commento:

Posta un commento