Tweet Geolocation Error Estimation

Part of:
Thursday Sep 29   03:30 PM to 05:30 PM (2 hours)
Twitter Location Error calculation
Tweet location is important for researchers who study real-time human activity. However, few studies have examined the reliability of social media user-supplied location and description in- formation, and most who do use highly disparate measurements of accuracy. We examined the accuracy of predicting Tweet origin locations based on these features, and found an average ac- curacy of 1941 km. We created a machine learning regressor to evaluate the predictive accuracy of the textual content of these fields, and obtained an average accuracy of 256 km. In a dataset of 325788 tweets over eight days, we obtained city-level accuracy for approximately 29% of users based only on their location field. We describe a new method of measuring location accuracy.