If you always kind of thought that Twitter is faster than authorities in accurately identifying threats, you're not that far from the truth.

New research from Cardiff University, which analysed 1.6 million tweets from the London 2011 riots, has noticed that the micro-blogging platform can be used to detect dangerous situations up to an hour faster than police reports.

To do so, researchers created event detection algorithms that use various features of Twitter data -- like sentiment, frequency of tweets containing certain words, and geolocation and timing of the tweets -- to cluster "similar" content and produce interpretable summaries.

Each cluster has variables like location, time sent, and hashtags used, and was given a timestamp.

"They [the algorithms] work by moving a sliding time window over the clusters so the content changes over time, enabling us to detect events over short periods," Dr. Pete Burnap, a Cardiff University researcher and co-author of the paper, said.

Surprisingly, perhaps, researchers discovered they were able to map out real-time disruptive events from five minutes to an hour before the Met Police were aware of them.

This example, during the incidents in Enfield Borough on Aug. 7, 2011, shows that the tweet about a fire and window-smashing in the area came more than one hour before the police report (the research paper contains only the content of the tweet to protect the individual's privacy):

Tweet may have been deleted

Barnap argues this prediction element could be used to help first responders and emergency services before the threat escalates.

"We argue that people are starting to self report incidents that they witness via social media quicker than people are phoning 999 in some cases, so in being able to cluster and summarise these self-reports we can augment traditional event detection methods so help the policing services," he said.

"Police are having to do the same or more in terms of event management, but on a decreasing budget."

A 'Minority Report' scenario?

However, there is an argument that "predicting" some kind of activity before it escalates could lead to a Minority Report-kind of scenario, in which law enforcement intervenes to curb freedoms of assembly or interfere with freedom of speech.

Last year, Twitter shut off commercial access to social media surveillance company Geofeedia after the ACLU revealed that Twitter, Facebook, and Instagram data was being marketed by Geofeedia to police as a way to monitor activists and protests.

The company told police they could track hashtags associated with Black Lives Matter protests, even using facial recognition software in real time.

Other surveillance programs include Beware, which markets itself as a service that can create "awareness of potential threats" and can assign a "threat level" to a social media user.

Media Sonar, whose website says it has "proven" to be a "powerful tool for assessing threats," was found to market their services to the Fresno Police Department, according to results of an investigation by the ACLU published in December 2015. 

Protecting privacy

Burnap recognises social media could be used to invade people's privacy and crack down on dissent.

"There must be stringent ethical practices in place to avoid this," he said, noting that his research works at an aggregate level, not pinpointing any specific Twitter users like Geofeedia and others did.

"Our research works at an aggregate level so really only flags events once there is sufficient ‘noise’ around a certain event," he says.

"This works in two ways -- firstly by avoiding identifying individuals in a effort to retain privacy, and also in that it requires effort and collusion to create fake reports as we’re not dependent on taking a single report as evidence of an event."

For instance, the research paper doesn't identify the individuals behind the tweets (Mashablewas able to uncover it using Twitter search).

The research could allow online observations of offline events, enabling better insights into public reports of disruptive events, Burnap says.

"The idea is that this research could augment more traditional approaches to event detection and sit alongside those traditional practices," he says.

So we're not in Minority Reportterritory, yet, but surely authorities are taking note. You've been warned.