Emergency Situation Awareness from Twitter for Crisis Management WWW 2012 Workshop on Social Web for Disaster Management Mark Cameron, Robert Power, Bella Robinson and Jessie Yin 17th April 2012 CSIRO ICT CENTRE Introduction The Australian Government’s Crisis Coordination Centre (CCC) • The CCC is a 24/7 all-hazards management facility • The CCC monitors domestic and international media for developments relating to national security and emergency management • When local authorities are unable to cope with a large scale emergency or disaster, the CCC coordinates assistance • The core operational processes of the CCC require verifiable facts Social media provides a new source of data from which crisis coordinators can obtain awareness of developing situations During the 2009 Victorian bushfires, situation information was reported in realtime on social networks, but this was not easily consumable by crisis coordinators ESA from Twitter for Crisis Management - Page 2 The Problem How can we help crisis coordinators extract useful information from Twitter? We have developed tools to: detect and alert on events of interest in near real time condense and summarise content to reduce watch officer load classify and review high-value messages enabling officers to quickly understand impact of real-world events on people and infrastructure focus and track issues as they arise, evolve and decay over time (still under development) perform forensic analysis of incidents ESA from Twitter for Crisis Management - Page 3 Architecture ESA from Twitter for Crisis Management - Page 4 Tweet Map showing captures and volume Red circles show capture regions Markers show geo-coded locations of tweets Number of tweets in 5 minute window ESA from Twitter for Crisis Management - Page 5 Near Real-Time Search • If terms used in an incident of interest are known, near real-time search techniques can be used. ESA from Twitter for Crisis Management - Page 6 Detecting incidents • When you don't know what to look for, such as with unexpected incidents, our Alert Monitor can provide clues to what is going on in Twitter Australia (and NZ!) wide. Our burst detector method examines the stemmed words in the tweets We’ve used historical data to build a statistical model of word occurrences A burst is defined as a positive variation from this statistical model ESA from Twitter for Crisis Management - Page 7 Condense and Summarise Messages • If an alert looks interesting we need to provide a way for rapid understanding of topics within bursting words • We have integrated a third party clustering engine Carrot2 to perform clustering of tweets • The cluster labels are displayed and the user can select a cluster to view the tweets belonging to it ESA from Twitter for Crisis Management - Page 8 Classify and review high-value messages • Our infrastructure impact classifier can identify Tweets that indicate damage to roads, bridges, power and telecommunications infrastructure. ESA from Twitter for Crisis Management - Page 9 Classify and review high-value messages (2) • Cluster labels for all Tweets immediately after the earthquake hit. • Cluster labels for Tweets classified as containing infrastructure damage immediately after the earthquake hit. ESA from Twitter for Crisis Management - Page 10 Forensic Analysis – Historical Alert Viewer • Christchurch Earthquake 4th Sept 2010 @ 02:38 (Aus time) ESA from Twitter for Crisis Management - Page 11 Forensic Analysis – Historical Alert Viewer (2) • Christchurch Earthquake 4th Sept 2010 @ 02:41 (Aus time) ESA from Twitter for Crisis Management - Page 12 Forensic Analysis – Historical Alert Viewer (3) • Christchurch Earthquake 4th Sept 2010 @ 02:57 (Aus time) ESA from Twitter for Crisis Management - Page 13 Deployment Experience • A subset of our tools have been deployed for trial with the Attorney-General’s Department, in particular the Alert Monitor and Historical Alert Viewer interfaces • We deployed this as an externally visible web application on our infrastructure (rather than setup the whole system on theirs) • In order to not violate Twitter’s Developer Rules of the Road, we needed to re-implement some of the system to re-retrieve tweet content directly from Twitter in order to display them (rather than simply get them from our database) ESA from Twitter for Crisis Management - Page 14 Deployed Web Interface ESA from Twitter for Crisis Management - Page 15 Future Work • Evaluate other burst detection techniques • Develop more classifiers • Carry out more trails with other government agencies ESA from Twitter for Crisis Management - Page 16 Thank you ICT Centre Bella Robinson Senior Software Engineer e bella.robinson@csiro.au w www.csiro.au CSIRO ICT CENTRE