Modification of the k-MXT Algorithm and Its Application to the Geotagged Data Clustering

摘要

The paper considers the problem of detection of the most attractive city sights using datasets with geotagged photographs. We form a graph on the basis of the geotagged spot coordinates and rewrite the problem as the problem of graph clustering. In this paper, we propose a modification of the k-MXT algorithm, which we called the k-MXT-W algorithm and which uses window functions. We compare the proposed algorithm with k-Means and k-MXT algorithms on simulated data using ARI, one of the most common metrics for assessing clustering quality. In this paper we also use the k-MXT-W algorithm to find the most popular places in St. Petersburg (Russia) and we compare the performance of the proposed algorithm with the k-MXT algorithm on real-world data using the modularity metric that does not require knowledge of true clustering.