Early Warning Tweets
Some researchers in Japan are turning to Twitter to help save lives.
They're using tweets to find the early warning signs of natural disasters. And they're finding benefits in the firsthand accounts published by users of the micro-blogging service.
A massive landslide struck a town in Nagano Prefecture in July. It killed a junior-high school student, and destroyed or damaged over 40 buildings and homes.
Twenty minutes before the disaster, someone posted on Twitter, "It's raining like mad. The river is cloudy."
The tweet was written by a woman named Eri Ito. "I was driving, and it was raining so hard that the rain in front of me was like a wall of white," she recalls. "It freaked me out."
What she saw was a sign of a coming landslide, but she didn't think much of it at the time. Her tweet was just like any other she has posted.
Researchers want to make use of tweets like the one posted by Ito. The Land Ministry and a private research institute have formed a team to develop a system that can find signs of impending landslides by analyzing tweets.
The team began by studying tweets made during a storm that cause a landslide in southern Japan 2 years ago.
A number of posts on the day include the phrase "the ground is rumbling," indicating the writers were facing imminent risk. The researchers hope to develop a system that will make the community share that sense of danger and urgency.
Masaru Kunitomo, who joined the team from the National Institute for Land and Infrastructure Management, says the warning signs were published on Twitter almost immediately. "Tweets describing ominous rumblings appeared at the same time as reports of noises from the ground and mountains," he says. "I think it will be useful to collect such information and use it as an early warning sign. It could help us evacuate people."
A computer system will scan tweets for keywords that could indicate landslides, such as "ground rumbling." Alerts would then be sent to local governments, as well as residents.
One challenge for the researchers is deciding on the keywords. A Japanese word that means something like "freaky", or "extreme" was used often on the day of the disaster. But it can be used in other situations. "On its own, a phrase like, 'Hiroshima is getting freaky,' doesn't indicate a coming disaster," a member of the team explains. "The question is how we interpret such language."
Another factor is the credibility of tweets, but the team finds the raw data beneficial. "This information is very useful as a starting point," says Atsushi Tanaka, director of the Center for Integrated Disaster Information Research. "It can help us understand what is going on, where it's happening, and what the dangers might be."
The government is funding the project for 2 year. The team hopes that will be enough time to unlock the potential of tweets as an early warning tool.