Augmenting mobile localization with activities and common sense knowledge

Location is a key element for ambient intelligence services. Due to GPS inaccuracies, inferring high level information (i.e., being at home, at work, in a restaurant) from geographic coordinates in still non trivial. In this paper we use information about activities being performed by the user to improve location recognition accuracy. Unlike traditional methods, relations between locations and activities are not extracted from training data but from an external commonsense knowledge base. Our approach maps location and activity labels to concepts organized within the ConceptNet network. Then, it verifies their commonsense proximity by implementing a bio-inspired greedy algorithm. Experimental results show a sharp increase in localization accuracy.