A new study by Cardiff University has determined that Twitter can be used to identify dangerous situations up to an hour faster than police reports.
Using a dataset of 1.6 million tweets from the London riots in 2011, researchers were able to have a series of machine learning algorithms automatically scan Twitter to identify threats. The system took into account things like the location of the tweet, the frequency of tweets containing certain words (or variants of these words), and timing of the tweets.
Applying these algorithms to the London riots, they were able to detect incidents faster than police in almost every instance. For example, their system’s summarization of “not feeling the rumors that the rioters are looking to move to edmonton and #enfield town. DON’T YOU PEOPLE THINK YOU’VE DONE ENOUGH!!!!” was picked up on 28 minutes faster than police intelligence of “information that known gang members were discussing moving onto Edmonton to cause disorder.”
They were able to detect incidents faster than police in almost every instance
As the report notes, many existing approaches to event detection are directed toward large-scale events like terror attacks, and it’s much harder to be alerted to smaller incidents like fires or car accidents. Leveraging social media data can solve this gap and also be applied to large-scale events, as well.
Not only could this method augment intelligence-gathering techniques already used by police, but picking up on these smaller events could help predict things like riots before they actually happen. Researchers said their “results show that our system can perform as well as terrestrial sources at detecting events related to the [London] riots; in some cases, we detect the event before intelligence reports were recorded.”
Cardiff’s report confirms that what companies like Dataminr have already been doing for governments and law enforcement clients works: aggregating what people broadcast on social media and turning it into real-time alerts for high-impact events.