Do regional boundaries defined by governments respect the more natural ways that people interact across space? This paper proposes a novel, fine-grained approach to regional delineation, based on analyzing networks of billions of individual human transactions. Given a geographical area and some measure of the strength of links between its inhabitants, we show how to partition the area into smaller, non-overlapping regions while minimizing the disruption to each person's links. We tested our method on the largest non-Internet human network, inferred from a large telecommunications database in Great Britain. Our partitioning algorithm yields geographically cohesive regions that correspond remarkably well with administrative regions, while unveiling unexpected spatial structures that had previously only been hypothesized in the literature. We also quantify the effects of partitioning, showing for instance that the effects of a possible secession of Wales from Great Britain would be twice as disruptive for the human network than that of Scotland.
Several attempts have already been made to use telecommunications networks for urban research, but the datasets employed have typically been neither dynamic nor fine grained. Against this research backdrop the mobile phone network offers a compelling compromise between these extremes: it is both highly mobile and yet still localisable in space. Moreover, the mobile phone's enormous and enthusiastic adoption across most socioeconomic strata makes it a uniquely useful tool for conducting large-scale, representative behavioural research. In this paper we attempt to connect telecoms usage data from Telecom Italia Mobile (TIM) to a geography of human activity derived from data on commercial premises advertised through Pagine Gialle, the Italian ‘Yellow Pages’. We then employ eigendecomposition—a process similar to factoring but suitable for this complex dataset—to identify and extract recurring patterns of mobile phone usage. The resulting eigenplaces support the computational and comparative analysis of space through the lens of telecommuniations usage and enhance our understanding of the city as a ‘space of flows’.
Recent developments in the field of machine learning offer new ways of modelling complex socio-spatial processes, allowing us to make predictions about how and where they might manifest in the future. Drawing on earlier empirical and theoretical attempts to understand gentrification and urban change, this paper shows it is possible to analyse existing patterns and processes of neighbourhood change to identify areas likely to experience change in the future. This is evidenced through an analysis of socio-economic transition in London neighbourhoods (based on 2001 and 2011 Census variables) which is used to predict those areas most likely to demonstrate ‘uplift’ or ‘decline’ by 2021. The paper concludes with a discussion of the implications of such modelling for the understanding of gentrification processes, noting that if qualitative work on gentrification and neighbourhood change is to offer more than a rigorous post-mortem then intensive, qualitative case studies must be confronted with – and complemented by – predictions stemming from other, more extensive approaches. As a demonstration of the capabilities of machine learning, this paper underlines the continuing value of quantitative approaches in understanding complex urban processes such as gentrification.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.