Location-based social networks (LBSN) are capturing large amount of data related to whereabouts of their users. This has become a social phenomenon, that is changing the normal communication means and it opens new research perspectives on how to compute descriptive models out of this collection of geo-spatial data. In this paper, we propose a methodology for clustering location-based information in order to provide first glance summaries of geographic areas. The summaries are a composition of fingerprints, each being a cluster, generated by a new subspace clustering algorithm, named , that is proposed in this paper. The algorithm is parameter-less: it automatically recognizes areas with homogeneous density of similar points of interest and provides clusters with a rich characterization in terms of the representative categories. We measure the validity of the generated clusters using both a qualitative and a quantitative evaluation. In the former, we benchmark the results of our methodology over an existing gold standard, and we compare the achieved results against two baselines. We then further validate the generated clusters using a quantitative analysis, over the same gold standard and a new geographic extent, using statistical validation measures. Results of the qualitative and quantitative experiments show the robustness of our approach in creating geographic clusters which are significant both for humans (holding a F-measure of 88.98% over the gold standard) and from a statistical point of view.
Shaping city neighborhoods leveraging crowd sensors
Information Systems, Elsevier, Vol.64, March 2017
© Elsevier. Personal use of this material is permitted. The definitive version of this paper was published in Information Systems, Elsevier, Vol.64, March 2017 and is available at : http://dx.doi.org/10.1016/j.is.2016.06.009
PERMALINK : https://www.eurecom.fr/publication/4957