Watch what you tweet

New Delhi, July 8: A software tool developed in India may help puncture the prominence of rumour-mongers on Twitter.

Computer scientists here have developed a software program that can assign scores to the credibility of Twitter messages and has drawn global attention as a tool to filter fake messages.

The program, developed through a three-year effort by researchers at the Indraprastha Institute of Information Technology, New Delhi, takes about six seconds to assign a tweet a credibility score.

Although still experimental, the program has drawn users from across the world - firefighters in Belgium, emergency response units in the US, and information technology research groups in academic institutions. The development comes amid concerns over the spread of false information through social media platforms such as Twitter, a web service with over 500 million users worldwide.

"Fake messages can mislead or even hurt people and cause financial loss," Ponnurangam Kumaraguru, professor of computer science at the IIIT who supervised the research effort, told The Telegraph. "But there is evidence to show that during major events, rumours spread faster than genuine information in the first eight hours."

TweetCred scrutinises tweets using a set of 45 parameters and assigns scores between 1 and 7 where 1 stands for low credibility and 7 for high credibility.

Aditi Gupta, a research scholar at the IIIT, says she was inspired to find ways to rate tweets after learning about rumours that spread through Twitter after the July 2011 Mumbai blasts. One tweet had falsely claimed a hospital needed a specific blood group, while another falsely claimed a girl had died, even posting a link of her picture.

Tweets after other events highlighted the urgency of the research effort. After the bomb blasts during the Boston marathon in 2013, a false tweet claiming an eight-year-old boy had died was retweeted over 30,000 times. Six false rumours after the incident were retweeted 130,690 times.

Most of the 45 parameters used to rate the credibility of tweets examine the user profile and the content from various angles.

"A tweet from me on the subject of Bollywood would have relatively low credibility compared to a tweet from Amitabh Bachchan," Kumaraguru said. "We use a number of such rules to calculate the probability of low or high credibility."

The researchers analysed over three million tweets to construct the algorithm that combines the analysis of each of the 45 parameters through a mathematical function to assign the score. "A low number of happy emoticons, a high number of sad emoticons, the inclusion of a URL (a website address) in a tweet, each of these parameters adds credibility," Gupta said.

Over the past year, several hundred Twitter users across Asia, Europe and North America have downloaded and installed TweetCred. The IIIT group estimated that cumulatively about 14 million tweets had been assigned scores.

Some scientists view TweetCred as a rare effort by an Indian academic institution to connect computer science with social science. "The problem of people spreading rumours through social media at a time of crisis is particularly serious for India where false information may trigger tensions between communities," said Dheeraj Sanghi, professor of computer science at IIT Kanpur, who was not associated with the research.

But researchers also caution that people wishing to spread false information and aware of the tool's credibility detection strategies may change their writing styles to gain credibility points. "The computer science community needs to constantly battle this menace," Sanghi said.