Uncovering spatiotemporal biases in place-based social sensing
Places can be characterized by the ways that people interact with them, such as the times of day certain place types are frequented, or how place combinations contribute to urban structure. Intuitively, schools are most visited during work day mornings and afternoons, and are more likely to be near a recreation center than a nightclub. These temporal and spatial signatures are so specific that they can often be used to categorize a particular place solely by its interaction patterns. Today, numerous commercial datasets and services are used to access required information about places, social interaction, news, and so forth. As these datasets contain information about millions of the same places and the related services support tens of millions of users, one would expect that analysis performed on these datasets, e.g., to extract data signatures, would yield the same or similar results. Interestingly, this is not always the case. This has potentially far reaching consequences for researchers that use these datasets. In this work, we examine temporal and spatial signatures to explore the question of how the data acquiring cultures and interfaces employed by data providers such as Google and Foursquare, influence the final results. We approach this topic in terms of biases exhibited during service usage and data collection.