Harnessing the hashtag: using social media to detect crises

Researchers develop a Twitter-based sensor for detecting emergencies as they happen.

Social media often feels like a necessary evil. It is good for making plans, but critics say it has eroded our privacy, inflated our vanity, shortened our attention-spans, and made it impossible to go one day without having to look at a cat meme.

It needn’t be all bad, though. It can be used to help people in crisis by detecting emergencies as they happen, a new study suggests.

Researchers Marco Avvenuti and Mario Cimino of the University of Pisa worked with Stefano Cresci, Andrea Marchetti and Maurizio Tesconi from the Institute of Informatics and Telematics developed a system that uses Twitter to identify emerging crises in “A framework for detecting unfolding emergencies using humans as sensors,” published in SpringerPlus on January 19, 2016.

With hundreds of millions of users worldwide writing short, topical and publicly available tweets, Twitter is an effective platform for the “human sensor.” The data, however, is very messy. The team’s challenge was to create a software system that extracts only the relevant information.

First, the system searches all tweets for emergency-specific terms like “earthquake” or “flooding,” and then filters them to get only those from eyewitnesses before removing any from a blacklist of untrusted users, such as hoaxers, trolls and fake accounts.

The system then performs a final fine-grained filter based on properties the team found to be characteristic of past tweets about emergencies—e.g., having fewer words, less punctuation and more slang. Emergencies are then detected by the increasing frequency of messages that pass all the filters.

Detection delay of the system versus INGV notification delays
Detection delay of the system versus INGV notification delays

The team validated their system using previous data on seismic events and found that the precision—the ratio of correctly detected events to total detected events—was 36% for earthquakes of magnitude between 2.0 and 2.0, 75% for those between 3.5 and 4.0, and 100% for those above 4.0. They also found their system was significantly more responsive than the National Institute of Geophysics and Volcanology (INGV), which monitors seismic events in Italy.

Although further work is required to validate the system on different types of emergencies, the results are certainly promising. It seems there may finally be something better to do with Twitter than follow the latest celebrity feud.

View the latest posts on the SpringerOpen blog homepage