No doubt social media is a great platform to spread news, data and even disaster alerts, but it has also become a medium to spread fake news and rumors – even fake disaster news that causes chaos among people. So to battle this issue, a bunch of students from IIT-Kharagpur developed an AI algorithm which will filter fake disaster news on social media and avoid any misunderstandings or unnecessary chaos.

The smart AI algorithm will basically extract critical information from social media platforms, which is hard to obtain manually and then run an authenticity check for such posts to decide whether it is real or fake news. If the news is found authentic, the AI algorithm will automatically pass on that data to relevant aid rescue and relief teams. Saptarshi Ghosh of the Department of Computer Science & Engineering said, “Our solution can detect fake news and can even alert users in the time of disasters through deep machine learning algorithms”.

The developers say that this algorithm is can distinguish fake disaster news from the real disaster news with at least 90% accuracy. And since only 2% of Tweets were found to have relevant information for a news on disasters, and most of the remaining are Tweets are just conversations about the disaster and sympathy for victims. However, the technique is still limited to just disasters, but using proper analysis and mapping received from the reliable sources, this can be expanded to identify unauthorized handles spreading general fake news.

Currently, a collaboration proposal has been sent to Microsoft Research and QCRI,  and once that is accepted, the IIT team will develop the systems for aiding post-disaster relief operations. That will give wings to the project that was initiated during the terrific Nepal earthquake in 2015 and the Chennai floods in 2016 – supported by funds from IIT Kharagpur’s Institute Scheme for Innovative Research and Development grant, Microsoft Research India and ITRA, Media Labs Asia and Department of Electronics and Information Technology jointly.

 

2