Authors: Morteza Karimzadeh*, University of Colorado, Boulder, Luke S Snyder, Purdue University, David S Ebert, Purdue University
Topics: Geographic Information Science and Systems, Cyberinfrastructure, Hazards, Risks, and Disasters
Keywords: Geolocation prediction, twitter, deep learning, geocoding, situational awareness
Session Type: Paper
Start / End Time: 11:10 AM / 12:25 PM
Room: Virtual Track 9
Presentation File: No File Uploaded
Real-time tweets can provide useful information on evolving events and situations. Geotagged tweets are especially useful, as they allow determining the location of origin and provide geographic context to online discourse. However, only a small portion of tweets are geotagged, limiting their use for situational awareness. To increase the amount of geotagged tweets, researchers have developed various algorithms for predicting the location of a tweet, such as deep learning classifiers and gazetteer-based methods. However, Twitter’s public feed has undergone changes that may affect such algorithms. Further, the utilization of state-of-the-art prediction models in real-time visual analytics systems has not been explored. In this paper, we report on our ongoing research and (i) adapt an existing geolocation prediction method, computationally improve its accuracy using word-level embeddings, and integrate it with SMART, a visual analytics system designed to facilitate situational awareness, (ii) demonstrate the utility of geolocation prediction for real-time systems, and (iii) evaluate the effectiveness of the integrated system using Twitter’s public feed effective in September 2019.