A University of the Western Cape computer science student has shown that Twitter posts can be used to develop a map that pinpoints hijacking hotspots, and now he’s planning to turn it into an app for the general public.

Taahir Patel, who presented his findings at Telkom’s Satnac conference in George on Tuesday, said he hopes his work will help mitigate the risk of hijacking in South Africa by drawing on social media for crowdsourced insights.

“A social media platform like Twitter offers a constant flow of real-time information. Using this stream of up-to-date data, it’s possible to plot visibly the occurrence of hijackings on a map, that can then be used to inform other users, notify emergency responders and even help law enforcement create incident reports,” Patel said.

“The trick is knowing what Twitter data is relevant… Aside from its character limit, Twitter doesn’t have many restrictions, so it’s crucial that we put the right parameters in place to extract valid data.”

To do this, Patel tested a combination of three machine-learning techniques – “multilayer feed-forward neural network” (MLFNN), “convolutional neural network” and “bidirectional encoder representations from transformers” – to separate the topic of “hijacking” from actual incident reports on Twitter.

“The MLFNN technique achieved 98.99% accuracy in determining the validity of tweets as hijacking reports. This is an extremely encouraging result and proves that we can successfully use social media data to develop a map that indicates hijacking hotspots,” said Patel.

The opportunities to glean other insights from social media are “endless”, he said. Work on developing a mobile app to “inform the general public of the dangers on the road” is ongoing.