DEALER: Distributed Clustering with Local Direction Centrality and Density Measure

Clustering by Measuring Local Direction Centrality (CDC) is a recently proposed innovative clustering method. It identifies clusters by assessing the direction centrality of data points, i.e., the distribution of their k-nearest neighbors. Although CDC has shown promising results, it still faces challenges in terms of both effectiveness and efficiency. In this paper, we propose a novel algorithm, Distributed Clustering with Local Direction Centrality and Density Measure (DEALER). DEALER addresses the problem of weak connectivity by using a well-designed hybrid metric of direction centrality and density. In contrast to traditional density-based methods, this metric does not require a user-specified neighborhood radius, thus alleviating the parameter-setting burden on the user. Further, we propose a distributed clustering technique empowered by z-value filtering, which significantly reduces the cost of k-nearest neighbor computations in the direction centrality metric, lowering the time complexity from O(n2) to O(nlogn). Extensive experiments on both real and synthetic datasets validate the effectiveness and efficiency of our proposed DEALER algorithm.