LocBERT: Improving Social Media User Location Prediction Using Fine-Tuned BERT

Abstract

Predicting user locations on social media platforms like Twitter is a challenging task with numerous applications in marketing, politics, and disaster response. This paper introduces LocBERT, a fine-tuned BERT model designed to accurately predict the locations of Twitter users based on their conversations. Our experiments focus on the “StateElecTweets” dataset, comprising 1.6 million labeled tweets associated with state locations within the United States. The results demonstrate that LocBERT outperforms traditional machine learning models such as SVM and Naive Bayesian, achieving an accuracy of 0.988 and an F1-score of 0.987. The study contributes to predicting the location of users in election-related tweets, enabling a better understanding of campaign demographics and assisting political stakeholders in refining their strategies. The findings of this research hold significant implications for various domains and highlight the effectiveness of LocBERT in accurately predicting user locations on social media platforms.