Your Actions Tell Where You Are: Uncovering Twitter Users in a Metropolitan Area

TL;DR

This paper introduces LocInfer, a system that effectively identifies the majority of Twitter users in a specific metropolitan area by leveraging geographic locality in user communications, achieving high accuracy with minimal data.

Contribution

LocInfer is a novel lightweight system that uncovers most users in a geographic area without scanning the entire Twitter network, validated through large-scale experiments.

Findings

LocInfer discovers on average 86.6% of users in a metropolitan area.

Achieves 73.2% accuracy in user identification.

Effective countermeasure for location privacy concerns.

Abstract

Twitter is an extremely popular social networking platform. Most Twitter users do not disclose their locations due to privacy concerns. Although inferring the location of an individual Twitter user has been extensively studied, it is still missing to effectively find the majority of the users in a specific geographical area without scanning the whole Twittersphere, and obtaining these users will result in both positive and negative significance. In this paper, we propose LocInfer, a novel and lightweight system to tackle this problem. LocInfer explores the fact that user communications in Twitter exhibit strong geographic locality, which we validate through large-scale datasets. Based on the experiments from four representative metropolitan areas in U.S., LocInfer can discover on average 86.6% of the users with 73.2% accuracy in each area by only checking a small set of candidate users. We also present a countermeasure to the users highly sensitive to location privacy and show its efficacy by simulations.