SocioLens: Potentials and challenges of large-scale social media data to understand human behavior in cities

Abstract

Urban scholars can now investigate complex phenomena on a larger scale and with lower costs, thanks to the advancements in big data collection and analyses. Among these data sources, social media data has been argued to be very useful for understanding human behavior and opinions. However, despite the considerable efforts in gathering and analyzing this emerging data source and the intense critics of its poor representation and potential biases, rare efforts have been made to compare the results generated by social media data and those revealed by other research methods. My dissertation lays out a research framework to explore the potentials and limitations of large-scale social media data in capturing and understanding human behavior, compared to traditional fieldwork methods (e.g., observation). Focusing on park use behavior, an essential pathway linking the built environment with human well-being outcomes, I extract behavior metrics from social media data using state-of-the-art machine learning models, triangulate social media-based results via systematic fieldwork, investigate how and why the discrepancies emerge, and propose responsible ways to deal with them. This study aims to create a heuristic about how to appropriately apply new technology for the betterment of cities and society.