Using GPS to learn significant locations

D. Ashbrook and T. Starner. Using gps to learn significant locations and predict movement across multiple users. Personal Ubiquitous Computing, (7):275–286, 2003.

This article shows an interesting approach to predict the user movements registering the user’s position through a GPS. The starting point is that for any technology to be useful the system must have a knowledge of what the user will do, where or when s/he will do it and the reason for the action. –In my case I am looking for the description of the place rather than on the user’s behavior.– The approach they followed is interesting and exemplar, because it is build on simple assumption, which are mainly inferred conjectures: a person who stops for a couple of time at a certain places may have particular interests there, therefore the place is a location. [other examples: stopping time marks the end and starting point of a trip (Wolf); loss of GPS signal is a building (Marmasse)]. On the application side there are several ideas proposed by the authors: spatial to-do-list: associate a task-reminder at a particular location; another may be the task sharing with other peers or the proximity-help-optimization: a friend is near a place you cannot go to pick up something and you ask for help; find people with similar interests based on physical proximity of certain preferential zones (Social Net – Terry). The goal they started with was to find a technique that would automatically pick out patterns in the data that would normally mirror what we observe about human movement. Then they reconsidered to aim for providing some starting inputs into the system that eventually would give some results. Their initial definition of “place” is interesting: any logged GPS coordinate with an interval of time “t” between it and the previous one. In addition they also did something interesting for defining statistically the notion of location and sublocation: they did clustering using a variant of the k-means clustering algorithm. They also used user inputs to enhance the identity of the space for the extent to verify the outcomes of the algorithm with the user’s expectations. The experimental model, though, I found weak. The conclusions is that they proved the possibility to detect space structures but their goal was also to detect time structures. They used a system called GPSVis to log data.