Paper: A Stacking-based Approach to Twitter User Geolocation Prediction

ACL ID P13-4002
Title A Stacking-based Approach to Twitter User Geolocation Prediction
Venue Annual Meeting of the Association of Computational Linguistics
Session System Demonstration
Year 2013
Authors

We implement a city-level geolocation prediction system for Twitter users. The system infers a user?s location based on both tweet text and user-declared metadata using a stacking approach. We demon- strate that the stacking method substan- tially outperforms benchmark methods, achieving 49% accuracy on a benchmark dataset. We further evaluate our method on a recent crawl of Twitter data to in- vestigate the impact of temporal factors on model generalisation. Our results sug- gest that user-declared location metadata is more sensitive to temporal change than the text of Twitter messages. We also de- scribe two ways of accessing/demoing our system.