Enhancing Location Recommendation Through Proximity Indicators, Areal Descriptors, and Similarity Clusters

Location recommendation (LR) or rather location-based recommender systems (LBRS) are an integral part of modern location-based services (LBS). Most LR algorithms only focus on location-specific attributes when calculating recommendations, while completely ignoring the urban structure surrounding the locations. (In this paper we refer to a geographic coordinate (latitude and longitude) as position. Locations and places in contrast refer to physical entities e.g. a restaurant, a bus stop or a lake). This paper demonstrates how the urban structure can be modelled in LR calculations by using data from OpenStreetMap (OSM) and the location data itself. Based on these datasets, we present two approaches to extend the LR process by (1) including the urban structure in direct proximity of the location (Proximity Indicators and Areal Descriptors) and by (2) not only looking for individual locations but location clusters (Similarity Clusters). Thereby we acknowledge the complexity of a location, which can not be perceived as a detached entity. A location is part of a given urban structure and we need to include the parameters of this structure in our algorithms. A prototypical implementation compares locations from four major German cities: Berlin, Hamburg, Munich and Cologne and thereby highlights the applicability of the underlying data structures derived from OSM and the location data itself. We conclude by outlining the potential of the presented approaches in the context of LR as well as their relevancy for urban planning and neighboring disciplines.