Analytical Workbench for Integrated Social Media Geo-Inference

Abstract

In the realm of social media monitoring and analysis, the availability of location-based information is of pivotal importance to understand the spatial behavior of social media users. Especially in fields like disaster management and urban planning, such data holds huge value for analysts and decision makers alike. However, as only few posts and messages in platforms like Twitter are already provided with GPS-coordinates or geo-tags by the users, researchers have proposed various algorithmic and modeldriven means to infer this information from properties like the content, network, or geographic history of the users. Since many of these methods only focus on isolated features or specific models, this paper presents a comprehensive framework that allows to integrate, combine and compare multiple geo-inference schemes in a unified, standardized, and performance-optimized fashion. In addition to that, we present a visual interface, which offers an intuitive, real-time assessment of the accuracy of singular and combined methods as well as support in detecting and understanding possible anomalies.We demonstrate the usefulness and relevance of our approach in a comprehensive case study.