Comparing Geospatiality of Topics between Geotag- and Geoparsing-based Geolocations

Abstract

Geolocated social media data offers the opportunity to analyze text data spatially in a wide variety of contexts. Previous work has identified that the likelihood of texts to contain mentions of locations varies between topics, indicating differences in their geospatiality. Social media posts can be linked to geographic locations through two main approaches: geoparsing, which extracts geographic information for places mentioned in the text, and geotags, corresponding to geographic coordinates explicitly attached to posts. In this study, we examine a curated data set of both geotagged and non-geotagged tweets for several thousand of Nigerian Twitter users, to explore differences between geotagging-based and geoparsing-based geolocation approaches in topic representation, controlling for the effects of users and time. Our findings indicate that the two approaches yield data with similar proportions of location mentions, but the interaction between topic and geospatiality varies substantially for some topics. We conclude that the method chosen to geolocate social media data can impact the number of geolocated posts differently across topics. This should be considered in research involving the identification of geolocations from social media posts.