Modeling Mobility Patterns Uncertainty of Twitter Feeds During an Earthquake Using Spatio-Temporal Point Process Models

Abstract:

An important aspect of designing an earthquake evacuation model is identifying how humans react and move to safety during a disaster. Social media outlets such as Twitter are becoming a commonly used data source for modeling spatio-temporal evolution of human movement before and after an earthquake. Despite its fine resolution, Twitter data contains spatial uncertainty due to partial privacy metrics and tweets where location is only contextual. We propose a space-time point process model to represent spatial uncertainty pertaining to Twitter data for the 2019 Ridgecrest Earthquake. Proposed model explicitly models spatial uncertainty of tweets with and without geotags, using contextual information from the microblog. Point process models utilize natural language processing (NLP) techniques to model location uncertainty of geotagged tweets. Results show the disparity in mobility pattern scenarios during the Ridgecrest Earthquake under spatial uncertainty. The importance of geotagged tweets is highlighted with uncertainty maps where location is inferred from microblogs. Results indicate that uncertainty propagates substantially for earthquake evacuation planning due to uncertainty in human movement patterns inferred from Tweets.