Geocoding Without Geotags: A Text-based Approach for reddit

TL;DR

This paper presents a novel text-based geolocation inference method for Reddit, overcoming the lack of geotags by creating ground truth labels and demonstrating domain-specific models outperform cross-domain approaches.

Contribution

It introduces a heuristic schema for labeling Reddit users' locations and evaluates geolocation models trained on Reddit and Twitter data, highlighting the importance of domain-specific training.

Findings

Domain-specific models outperform cross-domain models.

Interest-group metadata improves geolocation accuracy.

Effective heuristic schema for Reddit user location labeling.

Abstract

In this paper, we introduce the first geolocation inference approach for reddit, a social media platform where user pseudonymity has thus far made supervised demographic inference difficult to implement and validate. In particular, we design a text-based heuristic schema to generate ground truth location labels for reddit users in the absence of explicitly geotagged data. After evaluating the accuracy of our labeling procedure, we train and test several geolocation inference models across our reddit data set and three benchmark Twitter geolocation data sets. Ultimately, we show that geolocation models trained and applied on the same domain substantially outperform models attempting to transfer training data across domains, even more so on reddit where platform-specific interest-group metadata can be used to improve inferences.