Sensing 'Socio-Spatio-Temporal' Activity Patterns from Location Sharing Services Data

Abstract

Context: Online social networks and the Web 2.0 technologies embedded in these sites are creating an environment in which individuals can communicate and share information in ways that were previously not possible. Such websites are providing an unprecedented and growing collection of data on individual behavior that is very rich in detail. This includes information on with whom, when and where people interact, and more generally, what their activity patterns look like in time and space, and even what their sentiment or preferences are at specific moments and locations. Knowledge Gaps: There is a burgeoning body of literature that draws upon social media data and more broadly information collected via mobile communications devices (e.g., cell phone trajectories) to model and understand particular aspects of human behavior, including mobility patterns and social and spatio-temporal interaction. Yet, very little of this research has examined how and to what extent the spatio-temporal activity patterns revealed by these new forms of data vary across metropolitan areas, especially after controlling for relevant city-specific characteristics such as the size, density, composition or demographic profile of a city. While we recognize that the study of space-time activity patterns itself is not new, there are some gaps in the literature that should be noted. First, most analyses have been confined to a select set of cities – i.e., those that have conducted travel diaries or activity-based surveys. Due to inconsistencies in the format and type of information collected from such surveys, comparative analyses are problematic. Second, few studies have looked explicitly at the simultaneous integration of space, time and social (inclusive of cyber socialization) interaction, and the complex mobility patterns that arise from this behavior. Lastly, unlike location sharing services data, the information provided by travel behavior surveys tend to capture only mobility patterns arising from the primary residents of a city and not the behavior of transient visitors to that location. Study Objectives: The primary objectives of this study are to 1). understand how and to what extent location sharing services data approximate regional spatio-temporal activity patterns 2). develop a set of network-based metrics for characterizing the centrality and disorder of such activities in a region, and, 3). conduct a cross-city comparison using these metrics and related indicators of mobility.Data: To carry out the proposed research, we intend to use location services data collected over a five month period in 2010-11 (Cheng et al., 2011). This data provides information on user check-ins, or more specifically, where individuals indicate they are at different times of the day and week. Additional details on each individual’s status within the social networks that they belong are also included in the dataset. Methodology: The study methodology draws heavily on techniques from social network analysis, although concepts form landscape ecology, physics and geography are also utilized to capture different aspects of regional activity patterns. To gain an understanding of the types of activities that location sharing services data capture, we first conduct a correlation analysis using sector-based establishment data from the U.S. Census County Business Patterns. Correlations are examined at the zip code level. Second, using a space-time bipartite network topology, we derive a set of measures that characterize the centrality and disorder (entropy) of activities in a region, and that further can be decomposed to examine the spatial distribution of these characteristics. With individual location data aggregated to grid cells and summarized according to regular time intervals, we apply the technique to two U.S. metropolitan areas: Atlanta and Chicago.Significance: Implications for travel demand forecasting, epidemiological and information diffusion modelling and abnormal crowd detection (e.g., through “burstiness” analysis) will be drawn from the study.