A geographical and content-based approach to prioritize relevant and reliable tweets for emergency management

ABSTRACT

Tweets posted by the general public during disaster events represent timely, up-to-date, and on-site data that may be useful for emergency responders. However, since Twitter data has been deemed to be unverifiable and untrustworthy, it is challenging to identify those reliable and relevant tweets that can inform emergency response operations. Although computational methods exist both to classify overwhelming amounts of tweets and to filter those relevant to emergency response, using contextual geographic information regarding the disaster event to filter tweets has been overlooked. We review the existing research on the quality of data contributed by the general public from a geographical perspective, and then propose an approach to prioritize tweets for emergency response based on their relevance and reliability. The novelty of the approach is twofold: a) the use of both authoritative data such as hazard-related information and on-the-ground reports provided by weather spotters and validated by the National Weather Service; and b) the fact that it leverages tweets content as well as their geographical context and location. Using Hurricane Harvey in 2017 as a case study, results show that by following the proposed approach 79% of tweets sent from post-identified flooded areas were classified as of high or medium relevance and reliability. This suggests that the proposed approach can provide an accurate prioritization of tweets to be used for real time emergency management.

KEYWORDS:

• Twitter
• emergency response
• weather spotters
• spatial analysis
• reliability
• data quality

Acknowledgements

We thank the anonymous reviewers and editor for their useful feedback and insights that improved the final version of this paper. We also thank Daniel R. Montello and Leila M. V. Carvalho for their helpful feedback on a previous draft of this manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data and code availability statement

Data for this study are available in the references included in the Data section. The R code that support the findings of this study are available through Github: https://github.com/marcemiraval/Harvey-Project/.

Supplementary material

Supplemental data for this article can be accessed here.