Mapping near-real-time power outages from social media

ABSTRACT

Social media, including Twitter, has become an important source for disaster response. Yet most studies focus on a very limited amount of geotagged data (approximately 1% of all tweets) while discarding a rich body of data that contains location expressions in text. Location information is crucial to understanding the impact of disasters, including where damage has occurred and where the people who need help are situated. In this paper, we propose a novel two-stage machine learning- and deep learning-based framework for power outage detection from Twitter. First, we apply a probabilistic classification model using bag-of-ngrams features to find true power outage tweets. Second, we implement a new deep learning method–bidirectional long short-term memory networks–to extract outage locations from text. Results show a promising classification accuracy (86%) in identifying true power outage tweets, and approximately 20 times more usable tweets can be located compared with simply relying on geotagged tweets. The method of identifying location names used in this paper does not require language- or domain-specific external resources such as gazetteers or handcrafted features, so it can be extended to other situational awareness analyzes and new applications.

KEYWORDS:

• Power outage mapping
• social media mining
• deep learning
• natural language processing
• named entity recognition
• location detection

Acknowledgments

This manuscript has been authored by UT-Battelle, LLC under Contract No. DE-AC05-00OR22725 with the US Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. The Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan). Also, we would like to thank to Laurie Varma from ORNL for her diligent technical editing and proofreading of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Jibonananda Sanyal http://orcid.org/0000-0002-7789-3199

Budhendra Bhaduri http://orcid.org/0000-0003-1555-1377

Additional information

Funding

The authors would like to acknowledge the financial support received from Oak Ridge National Laboratory (ORNL)'s Liane Russell Distinguished Early Career Fellowship and grant no. TG0100000.