Characterizing Human Flows Extracted from Geotagged Social Media using Spatial Structural Similarity Index

Authors: Chanwoo Jin*, San Diego State University, Atsushi Nara, San Diego State University
Topics: Spatial Analysis & Modeling, Geographic Information Science and Systems, Hazards, Risks, and Disasters
Keywords: Social Media , Human Mobility, Evacuation
Session Type: Paper
Day: 4/13/2018
Start / End Time: 1:20 PM / 3:00 PM
Room: Bayside A, Sheraton, 4th Floor
Presentation File: No File Uploaded

Understanding human mobility is one of the long-discussed issues in geography, and it is important for planning of urban development, transportation, and evacuation. It has been challenging to measure dynamically changing human mobility due to lack of suitable tools and methods. However, recently the prevalence of real-time and human-created Big Data enables exploring various patterns of human movements. Especially, diverse social media allow us to track dynamic changes in human movements. However, few studies concentrate on comparing different mobility patterns of diverse media. Only some studies tried to measure differences with comparing their residuals or errors (Wu et al., 2014), which is less appropriate to deal with spatial context, spatial autocorrelation.
Therefore, this research suggests a new methodology to measure the similarity of different mobility patterns. It is a spatial version of structural similarity index (SSIM) with spatial weight matrix. SSIM is originally developed to assess images qualities (Wang et al., 2004). However applying SSIM directly to OD matrices causes a problem of sensitivity to OD pairs ordering due to the spatial autocorrelation of human mobility. In order to overcome the problem, this research suggests spatial weight matrix as a window in SSIM using pre-defined binary matrix by outer product with OD matrices. The spatial SSIM not only eliminates the sensitivity to OD pairs order but also enables to navigate local spatial patterns of similarity. It is expected to be utilized in evaluating estimated mobilities such as evacuation rates and comparing temporal dynamics of mobilities.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login