Bigfoot found? AI tool sifts fact from myth on Twitter
Posted: Thu Dec 19, 2013 9:35 am
- A matter of trust (Image: Thepalmer/Getty)
- mg22029484.700-1_300[1].jpg (16.74 KiB) Viewed 2313 times
Now Sejeong Kwon and colleagues at the Korea Advanced Institute of Science and Technology in Daejeon have designed an AI system that, they claim, does this correctly around 90 per cent of the time. If built into social networks, it could help us avoid embarrassing retweets or reshares.
The system analysed language used in more than 100 rumours – some later confirmed, others unfounded – that went viral on Twitter over a period of three and a half years. The researchers found that false rumours were far more likely to contain negative terms such as "no" or "not" than positive terms such as "like" or "love".
Being mentioned in "singleton" tweets – ones that were neither retweeted nor replied to – was another indicator of false content. Indeed, the best predictor that something was false was that it was tweeted separately by many users; accurate stories tended to have a few, widely retweeted sources.
The findings were presented at the International Conference on Data Mining in Dallas, Texas, earlier this month.
This article appeared in print under the headline "AI tool predicts when not to trust the Twitterati"