Best Paper Award

We were delighted to be awarded the Best Paper Award for our paper Extracting Information Nuggets from Disaster-Related Messages in Social Media. This described our recent work submitted to ISCRAM2013 where we explored the role of microblogging websites during emergencies. Microblogging websites, such as Twitter, play an important role during a crisis. When a disaster strikes, hundreds of thousands of microbloggers come to Twitter to share their experiences. This huge number of tweets contain informative messages that can be used for situational awareness, and not only that, the challenging part of it is that a large number of these messages may be misleading, non-informative hence not interest of crisis responders. However, the informative messages may contain key information such as damage reports, donations services, casualties reports, shelter announcements, food availability, which is very crucial for crisis responders to obtain and to accordingly act upon.

In this paper, we introduced an automatic system which extracts information from microblog posts. Specifically, we focus on extracting brief, self-contained actionable messages relevant to disaster response. We used real-time supervised learning methods combined with machine learning approaches to perform classification and information extraction tasks. For experimentation and evaluation we collected Joplin 2011 tornado dataset that struck Joplin, Missouri in the late afternoon of Sunday, May 22, 2011. The 206,764 unique tweets were selected by monitoring the Twitter Streaming API using the hashtag #joplin a few hours after the tornado hit.

We used CrowdFlower crowdsourcing platform to annotate 4,406 messages sampled uniformly at random from the dataset and asked workers to label using the ontology shown in the paper. A set of multi-label classifiers were trained using various binary as well as linguistic features to automatically classify a tweet into one or more of the identified classes. For this purpose, we use Naïve Bayesian classifiers as implemented in Weka. High performance measures (i.e., precision, recall and auc) prove the viability of our system.

The results of this applied and ongoing research will continue to inform the development of the Qatar Computing Research Institute’s (QCRI) Artificial Intelligence for Disaster Response (AIDR) platform. The purpose of this platform is to provide humanitarian organizations like the United Nations Office for the Coordination of Humanitarian Affairs (UN OCHA) with a platform that will enable them to create their own automatic classifiers on the fly.

See a follow-up to this work here which is published in WWW-2013.