Deducing places of interest from clusters of locations

Abstract: Some Location Based Services (LBS) can automatically find geographic locations that are relevant to the everyday smartphone user. A relevant location, or place, is a location that is of some significance to a user, e.g. home, workplace, airports or stores. Knowledge of these places can be used to enhance a smartphone application. However, most approaches to finding places are coarse, and simply define a place with a circle or polygon representing a geographical area. Instead this paper explored the feasibility of defining a place by using the natural boundaries found in the information of a map. The developed algorithm calculated the center of a cluster of location points by adding biased weights to each point. A close proximity of the center point was then searched for certain types of map elements such as buildings or parks. Because of time restrictions, map images were used instead of the underlying data. The developed algorithm found the correct place in 78% of the 45 test cases. In 15% of the cases it could not find anything, mainly because the map did not contain sufficiently detailed information about buildings outside of cities. The remaining 7% were incorrect results, some of which might have been remedied by more detailed map information. Overall the suggested approach was viable when a user had been in a building, park, or other clearly defined place, and when there was sufficiently detailed map information. To further this research an algorithm that processes geographical data directly instead of using map images could be tested. It would avoid some of the problems created by having an image as a middle layer between data and algorithm.