Activity Recommendations With Real-Time Location Data

The launch of Citysense, a real-time social navigation and nightlife discovery application for Blackberry and iPhone confirms the trend of analyzing massive amounts of real-time and historic location data from mobile devices for predictive analytics. The idea is similar to automobile GPS systems sharing and pooling current road speed conditions so that everyone can avoid congestion (see Real-Time Traffic Routing from the Comfort of Your Car). The algorithms behind Citysense indexes the active places in a city and characterized them by activity, versus proximity or demographics, to better understand the context of people behavior. In the same range, using the iPhone’s map and self-location features, as well as information about the prior activities of the user’s friends, Whrrl proposes new places to explore or activities to try. Last year, PARC developed, Magitti, a mobile application that uses a combination of cues to infer her interests. With the time of day, a person’s location, her past behaviors, and even her text messages the application suggests concerts, movies, bookstores, and restaurants.

The recent release of these social softwares is also a signal of a shift from answering “where are my friends” to “where is everybody” or “what is everybody doing” (A theme I discussed in the paper “Leveraging Urban Digital Footprints with Social Navigation and Seamful Design“). But they also raise the questios: are people interested in being entertained with real-time information and can machine learning algorithms provide solutions?

Citysense screenshot on a Blackberry

Relation to my thesis: Sense Networks, the company behind Citysense also developed Macrosense, a direct technology transfer from Sandy Pentland’s work on “Reality Mining” (see Nathan Eagle’s talk at Lift on the subject). It demonstrats the real interest in combining massive amounts of anonymous, aggregate location data to understand people dynamics (see Understanding Human Mobility Patterns). However, for powerful predictions there is still a lot of work to do in 1) understanding what motivates individuals to behave the way they do and 2) how users perceive and interact with the information and recommendation.