Inference of human mobility networks through the assimilation of social media traces into mobility models

MARIANO G. BEIRO; A PANISSON; C CATTUTO

Tempe

Conferencia; Conference on Complex Systems 2015; 2015

Complex Systems Society

The knowledge of high-resolution human mobility maps is important for a number of applications, such as transportation planning or the study of epidemics spreading. However, detailed empirical data on air travel and commuting have limited availability for some regions and countries in the world. Additionally, they are provided at a particular resolution level which may not be the one needed in a certain situation. For this reason, models that are able to infer information about human mobility networks are of much interest.On the other hand, social media are a promising source of information for studying socio-technical systems with computational tools. Online systems that collect georeferenced data are available almost worldwide, and they provide a high resolution description of mobility. In particular, data from services like Foursquare, Twitter and Flickr can be assimilated into mobility models in order to devise new models that show better agreement with empirical data.Here we use the Yahoo Flickr Creative Commons 100M dataset (consisting of 100 million worldwide taken photographs) for modeling human mobility flows. We develop a workflow for transforming the georeferenced traces at the user level into user itineraries formed by sequences of cities. Then, by aggregating all the users' itineraries at a given geographical resolution, we obtain a data-driven mobility model. We calibrate and validate our model using a dataset from the Bureau of Transportation Statistics of the United States. We show that a hybrid model that enriches the gravity model with mobility traces from Flickr can achieve a higher performance than the gravity model alone in terms of R^2 correlation coefficient. The methodology and results presented in this work pave the way for new hybrid approaches that combine mathematical mobility models with digital traces from social media.