Automated text processing often struggles with inaccurate geolocation mentions. This article introduces a two-stage supervised machine-learning algorithm designed to tackle this issue.
We extract contextual information from texts, including N-gram patterns for location words, mention frequency, and sentence context containing these words. Using training data sets, we estimate model parameters before applying it to test data sets.
Our results demonstrate that the algorithm significantly outperforms existing geocoders, even when handling additional zeros in a post-hoc case study. This advancement enhances event localization accuracy using news articles from around the world.
| Modeling Related Processes with an Excess of Zeros was authored by David Carlson. It was published by Cambridge in PSR&M in 2019. |
