Dynamic Spatial Cluster Process Model of Geo-Tagged Tweets in London

Abstract

Geo-tagged social media data is a key input to many smart city application areas, ranging from mapping consumer demand to understanding location dependent well-being. The sparsity in geo-tagged data, especially in certain cities, means that there is a lack of dynamic spatial point process models for social media data. Having statistically representative spatial models can enable proxy models that improve our understanding of human patterns in urban and suburban areas. Here, we analyse a data set of more than 400,000 Tweets in London to create a spatial point process model of Tweet clusters. We model Tweet clusters as a Poisson Cluster Process. We then track how the point process parameter and spatial entropy evolve over time to create a generative model usable for others, as well as discuss its relevance to urban dynamics and smart city applications.