Using social media to inform and help citizens during disasters
Social media outlets are beating traditional media outlets to the “in-the-moment” information punch. In the US, Hurricane Sandy news traveled faster over Twitter than on television. Live updates from the Arab Spring were on Facebook before traditional news outlets. To systemize these bursts of citizen journalism – and make them available and useful to disaster-affected communities during times of crisis – a team of scientists at IBM Research’s lab in Melbourne is using machine learning and text analytics to automate the collection and presentation of potentially critical information resource.
“Social media is a critical information source in emergency situations, creating the potential for every person on the ground to become a frontline journalist.”
– Jennifer Lai, IBM Research Manager for Intelligent Information Interaction
The Australia team had seen first-hand the limitations of relying solely on websites. The Victoria fires in 2009 were so severe – claiming 173 lives and injuring more than 5,000 people – that queries for updates and news brought down government websites dedicated to covering the disaster. Swinburne University of Technology reported in 2010 that 1,684 tweets about those fires “were laden with actionable factual information which contrasts with earlier claims that tweets are of no value made of mere random personal notes.”
In 2011, working with several universities, they launched the “OzCrisis Tracker” for mining social data including Twitter activity about bushfires, floods and other significant natural events across the continent.
But OzCrisis had limitations exacerbated by the growth in social media. Volunteers were curating tweets by categorizing them into logical events and placing those events on a map. That was no longer viable in a world where 20 million tweets were shared during Hurricane Sandy.
“We needed to eliminate the high dependency on this volunteer interaction and instead, augment the automation of the tool,” said Chris Butler, IBM Research scientist and project team lead.
Separating – and learning from – the noise of disasters
Fast forward to 2013 and an automated version of OzCrisis called Australia Crisis Tracker (ACT).
The team applied machine learning techniques on the data to filter spam and categorize event types, such as fires, floods, storms, and others – going well beyond hashtag hunting. ACT users log in with their twitter credentials, giving users easy access to the rest of Twitter. They can also click on an event to read the latest tweet reports from those on the ground, as well as view tweets from related events.
“While news agencies often place emphasis on hashtags, they can be unreliable when it comes to clustering and tracking events,” said Josh Andres, IBM Research scientist and project user experience designer.
ACT also automatically clusters tweets related to the same event, and places them on a map using geocoding techniques. Using Natural Language Processing, ACT augments the number of tweets that can be geocoded beyond the small number of tweets that have GPS coordinates automatically associated with them. Community and government agency tweets are also available in the tool.
Expanding ACT
Funded in part through a grant from IBM Corporate Citizenship and Corporate Affairs, the team deployed ACT with the Australian Red Cross’ Emergency operations Centre (EoC) for the state of Victoria, headquartered in Melbourne. An EoC situational awareness team, which also includes volunteers, uses ACT and other tools to create disaster intelligence reports for their commanders, who share the needs and voice of the community with the broader command and control center activated during disasters.
"Micro-blogging is a potentially rich source of information for those working to save lives after disasters, and Australia Crisis Tracker goes far in harnessing that potential,” said Diane Melley, Vice President of IBM Corporate Citizenship and Corporate Affairs. “The partnership with the Australian Red Cross was a good environment for proving the concept and envisioning where it can go from here."
The team’s next technical upgrade is to include tweeted photos in ACT’s information stream. And in preparation for using this solution in other countries for other types of disasters, such as mud slides and earthquakes, the team is working on creating a more general purpose geocoder and using IBM SoftLayer cloud technology.
No comments:
Post a Comment