Work

Twitter Geolocation

Year: 2018

Type: article

Abstract: Geotagging Twitter messages is an important tool for event detection and enrichment. Despite the availability of both social media content and user network information, these two features are generally utilized separately in the methodology. In this article, we create a hybrid method that uses Twitter content and network information jointly as model features. We use Gaussian mixture models to map the raw spatial distribution of the model features to a predicted field. This approach is scalable to large datasets and provides a natural representation of model confidence. Our method is tested against other approaches and we achieve greater prediction accuracy. The model also improves both precision and coverage.

Source: ACM Transactions on Knowledge Discovery from Data

DOI: 10.1145/3178112

Authors: Jordan Bakerman, Karl Pazdernik, Alyson Wilson, Geoffrey Fairchild, Rian Bahran

Institutions: North Carolina State University, Los Alamos National Laboratory

Language: English

Topic: Human Mobility and Location-Based Analysis

Subfield: Transportation

Field: Social Sciences

Domain: Social Sciences

Open Access status: green

Funders: National Science Foundation, U.S. Department of Energy, North Carolina State University, National Nuclear Security Administration

Awards: G1021341557, G1094073322, CONSORTIUM FOR NONPROLIFERATION- ENABLING CAPABILITIES (PDP14-04), Collaborative Research: NRT-DESE: Interdisciplinary Research Traineeships in Data-Enabled Science and Engineering of Atomic Structure