This article was taken from the January 2014 issue of Wired magazine. Be the first to read Wired's articles in print before they're posted online, and get your hands on loads of additional content by subscribing online.
When hurricane Sandy hit the US east coast, fake images quickly went viral. To help spot fakes, researchers at the Indraprastha Institute of Information Technology, Delhi, applied a machine learning system to tweets that contained an image URL and the words "hurricane" and "sandy". The algorithm did not analyse the images, but used data in the tweets to classify them. The study's lead author Aditi Gupta and her PhD advisor Ponnurangam Kumaraguru explain how.
Check the source
A tweeted image's origins say a lot about its authenticity. If it comes from a user who has been on Twitter for a while and has a lot of followers, it's more likely to be legitimate. "A lot of fake images were uploaded by people with fake accounts," says Gupta. If an account is created in honour of an event, its information is much less likely to be real. Retweets accounted for 86 per cent of the fake images that were shared, and 90 per cent of those RTs came from 0.3 per cent of users -- these influential tweeters are responsible for the images going viral.
Look at the language
The content of a tweet containing an image URL is a good indicator of its authenticity. "With a real image, we found the language was much more formal; the sentence structure was better," says Gupta.
Fakes are more likely to be accompanied by incomplete sentences containing lots of abbreviations, smileys and slang. The algorithm also took into account the length of the tweet, uppercase letters and punctuation marks -- but because multiple factors affected its classification process, it's not possible to say exactly how these features differ for real and fakes.
Use a web tool
The researchers have used their studies to create tools anyone can use. Their web app Twit-Digest lets you search for tweets on a specific trend, which it will then award a star rating for credibility. They have also just released a plugin that will comb your feed and flag unreliable posts in real time. "It can sit on your browser, and any time you go to Twitter it can look at the tweets on your timeline and say whether they are credible or not," explains Kumaragu. "We're trying to build technologies which can reduce the kind of virality of misinformation."
This article was originally published by WIRED UK
