Using tracking data from peoples’ phones, a team of British researchers are developing an algorithm that predicts where the user will be in 24 hours time. The prediction is surprisingly accurate, with the average error down to a mere 20 metres from the correct location after three hours. Now the researchers that developed it are planning to take tailored advertising to new levels. Businesses will be able to predict where consumers are going, and target them with location-specific offers. What’s more, they won’t even need to speak to the consumer in order to produce this personalised advertising.
As understanding of human behaviour increases, the methods of exploiting it for commercial gain are too. Earlier in the year, Target was revealed to be collecting and interpreting customer data so effectively that they can not only tell with a good degree of certainty when a customer is pregnant, but also when their due date is likely to be. Such technology, once perfected, will enable advertising on a person-by-person basis that suits the consumers’ needs as well as their schedule.
Most people follow consistent routines over time but breaks in those patterns have been hard to predict with accuracy. That has changed, though, with the recent research undertaken by Mirco Musolesi, Manlio Domenico, and Antonio Lima. The research combines location data from a participants’ phone with that of their contacts. By looking at the correlation between users they can forecast where a person will be going over the next day.
Musolesi has worked in the area of human mobility prediction for several years and leads a group at Birmingham University, working in large-scale data mining, networks, and systems. He previously developed a model of human mobility based on social networks: “One of the main goals of my research work is to devise models that are able to exploit the availability of the exponentially increasing amount of multi-dimensional information about users and their environment, such as location, personal profiles, social network data and so on.”
“In the study we consider pairs of users, such as a user and his friends. You might get an improvement if you consider more than one user but you might introduce noise if the movement patterns are not sufficiently correlated. The key problem is that if you consider more than two users, the underlying model gets more and more complex and a very large amount of data is then necessary to train it.” The researchers are currently working to resolve these issues.
This could lead to uses such as individual-specific advertising that targets consumers based on where they are likely to be at a future time: “Since we are able to predict the future location of a user, an application can use the algorithm to send adverts to users in the morning about lunchtime offers of restaurants in the area where a user will go.”
That offer could be further tailored based on what restaurants the consumer usually eats at, and what food they prefer. The ideas are still in infancy, and more uses are arising: “Other applications are non-commercial ones such as targeted information related to concerts or exhibitions in areas of the city at times where a user is likely to be in the future.”
“It might also be used to predict the future locations of events in general. For example you can apply the algorithm to predict the future location of a criminal hotspot.”
Musolesi thinks users should be allowed to opt-in rather than have their data automatically shared, citing privacy as one of his main concerns: “One possibility is to share mobility information among friends and use encryption for data sharing, especially of GPS locations.”
Consumers are already sharing location information through social networks and receive offers for doing so. Forecasting their movements is likely to be a blessing for both parties; customers get relevant and interesting offers while businesses get exposure.
Musolesi stated that they are thinking of providing a free, for the time being, API (Application Programming Interface) to third-party developers of apps. This would open the doors for all manner of industries wishing to predict the movements of smartphone users.
Over the coming months the researchers will be testing more refined techniques for prediction in order to reduce variance between results. Musolesi hopes the technology may be commercialy available within two years. Until then, consumers will have to decide for themselves where they want to eat lunch.