Skip to 0 minutes and 7 seconds The thing that you want is to try to predict the GPS coordinates of the individual. And the one thing that is quite interesting is that this is possible, but the problem is that you need data about that. And the data usually comes in the space from companies and we participated in the Nokia Mobile Data Challenge a couple of years ago. And we got data from 152 smartphones. Actually we got a smaller data set subset out of this big data set. And from this data, essentially we built a predictive model of human mobility that is also able to exploit correlation between the users.
Skip to 0 minutes and 55 seconds And this is the powerful thing when you use big data is the fact that you can use not just the data of a single user, but you can correlate data.
Skip to 1 minute and 7 seconds So you have the movement patterns of multiple people. We did a predictor that doesn’t just consider the movement pattern of a single user, but also other users that have mobility patterns that are highly correlated with this user. OK. So you have a set of traces. You can select the traces of users that are highly correlated in terms of mobility patterns. This is quite interesting because then you can view the predictor. So we used a multi-variate non-linear predictor that has an input, not just the input pattern of a single user, but also the mobility pattern of a different user.
Skip to 1 minute and 50 seconds And the interesting thing here is the fact that you can see also a correlation with the social links, social information that you get. Because we also got for these users, their address book. And with the address you can infer friendship. The fact that these people know each other. We considered them friends if they had each others phone number. Usually you have the phone number of your plumber. Your plumber doesn’t have your phone number. You might have a friend who is a plumber. So the idea is that we use this as a proxy. But we found a very interesting correlation between social links and correlation in terms of mobility patterns, at least in this data set.
Skip to 2 minutes and 37 seconds I don’t want to make strong claims about these things. So I want to show you a very short video.
Skip to 2 minutes and 47 seconds OK, start. OK. So this is the prediction with one user. So the flag is a predictor.
Skip to 2 minutes and 59 seconds And you see the area is quite large for one user.
Skip to 3 minutes and 8 seconds So the data was from the Nokia research lab in Lausanne.
Skip to 3 minutes and 19 seconds So this is just an example.
Skip to 3 minutes and 33 seconds And then if you use correlation, because you see that the users have synchronised– and we saw this by visualising the data, what you get essentially is a prediction at a block level after three hours and around one mile for all the users after 24 hours in this data set, if you consider social ties.
Using data from mobile phones to predict where people are going
We’ve seen that data from mobile phones can help us understand where people are. Can we use the same data to predict where people are going?
Mirco Musolesi presents new research on using smartphone data to predict users’ journeys. He explains how predictions become much better if you consider who mobile phone users are friends with.
Mirco Musolesi is a Reader in Networked Systems and Data Science at the School of Computer Science at the University of Birmingham. His research interests lie at the interface of ubiquitous computing, large-scale data mining, and network science.
You can watch the whole of Mirco’s presentation “Mining and Understanding Big (and Small) Mobile Data” on YouTube (21:05).
© Warwick Business School, The University of Warwick