Detecting emergencies using social media

Social networking has changed the way people broadcast and receive information. Every minute, vast amounts of information are communicated via Twitter. CSIRO has developed an analysis tool that helps organisations make sense of massive volumes of social media traffic. The software is providing solutions for a variety of applications including: management of emergencies, reputations and mental health.

The Emergency Situation Awareness (ESA) software detects unusual behaviour in the Twitter stream and quickly alerts the user when a disaster event is being broadcast. ESA also stores complete Twitter stream information and allows post-event analyses. Such useful and accessible information will provide timely situation awareness for disaster managers and emergency response agencies.

Responding earlier to bushfires, finding false claims for reputation management and helping address mental health issues are some of the benefits being seen with the introduction of new social media software from CSIRO. Without suitable tools, social media information can’t be used. For example, details about the 2009 Victorian bushfires were reported in real time on social network sites but were not visible to state or federal disaster response agencies.

Alan Dormer, CSIRO Services Science Leader, said that with millions of posts and countless conversations happening every minute, organisations trying to make sense of social media can easily find themselves overwhelmed.

“There are 11.5 million Facebook accounts in Australia and more than two million Twitter accounts, so analysing social media posts to find relevant information is like looking for a needle in a haystack,” Dormer said.

“It’s a classic big data problem. But with our research in data mining, textual analysis and data visualisation, we’re well placed to tackle it,” he said.

So far, organisations are using the software for three main reasons - reputation management, exploring topics and issues important to the community, and early detection of emergencies or outbreaks. And it’s showing promising signs of increasing efficiency and productivity.

Prof Allan Fels, Chair of the National Mental Health Commission, said he has found social media analysis gives his organisation insights into community thinking on mental health and wellbeing.

“We believe mental health and suicide prevention is an issue for all Australians. The CSIRO social media engagement tools help us identify key issues on a daily basis and provide social media reports which are easy to understand and quick to produce,” he says.

The commission plans to use the software to gauge community response to its report cards on Australia’s mental health issues and services. The first is due out later this year.

CSIRO used its own social media analysis tools to find false claims about coal seam gas research that appeared in social media. This allowed CSIRO to address the misinformation quickly.

Social media analysis in action

A hospital was recently threatened by a grass fire in Cloncurry, in outback Queensland. The CSIRO ESA software was used to analyse Twitter post and gave the Queensland Department of Community Safety early warning of the incident.

Tweets about the fire emerged well ahead of any official alerts; within minutes, details such as the fire’s location and direction were appearing on Twitter, allowing  emergency managers to evacuate the hospital safely.

ESA provided crisis coordinators early notification to prepare their response to the fire while waiting for confirmation from official channels. This meant the evacuation plan could be prepared, providing precious extra time to the emergency management workers on the ground.

It allowed efficient, safe and timely evacuation of hospital staff and patients before the fire got out of control and evacuation became difficult or impossible.

The potential applications of social media information for disaster managers include providing:

• evidence of pre-incident activity;
• near real-time notice of an incident occurring;
• first-hand reports of incident impacts; and
• gauging community response to an emergency warning.

How the software works

ESA exploits the statistical incidence of words used on Twitter to describe emergency events. It’s trained using historical word occurrences on Twitter from past disaster events. This allows ESA to reveal emerging topics and flag them for investigation. Searches repeated every minute look for words that are used more often than normal and these detected ‘word bursts’ are extracted and stored, and are available for access by incident response agencies via the ESA web application.

The ESA software can:

• detect unexpected or unusual incidents, possibly ahead of other communications;
• condense and summarise messages about an incident maintaining awareness of aggregated content without having to read individual messages;
• classify and review high-value messages during an incident (eg, messages describing infrastructure damage or cries for help); understand the impact of an incident on people and infrastructure;
• identify, track and manage issues within an incident as they arise, develop and conclude; proactively identify and manage issues that may last for hours, days or weeks; and
• perform analysis of incidents by exploring social media content from before, during and after an incident.

ESA provides situation awareness by using data mining techniques including burst detection, text classification, online clustering and geotagging. These techniques are adapted and optimised for dealing with real-time, high-volume text streams which identify early indicators of unexpected events, explore the impact of identified incidents and monitor the evolution of events.

Dormer said the social media analysis tools were being developed with government, for government - the Australian Government Department of Human Services being a key partner. Business is also starting to show interest.

“We’ve formed an ‘early adopters group’ of innovators in government to help us develop the social media tools beyond the prototype stage, trial them in real situations and give us feedback to make them more useful,” he said.