30 May 2022

In daily life, personal data about our movements and habits are constantly collected by smart devices carried on our person. The data collected by our mobile phones and fitness trackers not only tracks our physical movement, but can also leave movement traces embedded within our social networks.
Contrasting social and non-social sources of predictability in human mobility

A new study coming from the University of Edinburgh Business School in close collaboration with the Universities of Rochester, Exeter, Vermont, Rio de Janeiro and Northeastern, has analysed four datasets and developed a ‘colocation’ network framework to further investigate mobility behaviours. Three of the datasets took publicly available location-based data points from Brightkite, Facebook and Foursquare. The fourth dataset came from phone call record data collected in Rio de Janeiro Metropolitan Area (RJMA) in Brazil—comprising 22 million calls by 36,000 anonymous users.

"What we did in this analysis was use this ‘colocation’ network to compare mobility behaviours between two set of people", said Dr Zexun Chen, Lecturer in Predictive Analysis. "Firstly people who are socially tied to an individual (such as colleagues, friends, or relatives) and secondly people who are not socially tied to an individual, but who happen to visit a specific location at a similar time to another individual. A nice example of this is picturing two students who are unknown to each other, but who attend the same university and visit the library at the same time."

By applying entropic measures (the degree of information contained in a sequence of location visits), the research team found that the movement patterns of people socially tied to one another contain up to 94% of the information needed to predict that individual’s mobility patterns.

More surprisingly, the team then found that an aggregation of strangers (people not socially tied) could predict up to 85% of an individual’s movement.

"These results presented here constitute a double-edged sword in terms of its implications," said Dr Chen. "While these data can inform important applications such as contact tracing in the early stages of a disease outbreak, significant ethical concerns surrounding such information sources make it critical to place strong access constraints on mobility information."

This paper has furthered the impetus on the ongoing debate around best practices for privacy protection, both in terms of legislation and ethical algorithmic development. The team are now exploring methods of protecting personal mobility data through employing advanced privacy-preserving techniques. Further application of this research can be applied to smart urban planning and early pandemic control, and exploring questions like: how do human movements in urban areas affect the social-economic development?

To read more about the paper published in Nature Communications, visit: Contrasting social and non-social sources of predictability in human mobility.

Dr Zexun Chen is a Lecturer in Predictive Analytics.