Detecting country of residence from social media data: a comparison of methods

Abstract

Identifying users’ place of residence is an important step in many social media analysis workflows. Various techniques for detecting home locations from social media data have been proposed, but their reliability has rarely been validated using ground truth data. In this article, we compared commonly used spatial and Spatio-temporal methods to determine social media users’ country of residence. We applied diverse methods to a global data set of publicly shared geo-located Instagram posts from visitors to the Kruger National Park in South Africa. We evaluated the performance of each method using both individual-level expert assessment for a sample of users and aggregate-level official visitor statistics. Based on the individual-level assessment, a simple Spatio-temporal approach was the best-performed for detecting the country of residence. Results show why aggregate-level official statistics are not the best indicators for evaluating method performance. We also show how social media usage, such as the number of countries visited and posting activity over time, affect the performance of methods. In addition to a methodological contribution, this work contributes to the discussion about spatial and temporal biases in mobile big data.