Journey to the Centre of the Star: Various Ways of Finding Star Centers in Star Clustering

The Star algorithm is an effective and efficient algorithm for graph clustering. We propose a series of novel, yet simple, metrics for the selection of Star centers in the Star algorithm and its variants. We empirically study the performance of off-line, standard and extended, and on-line versions of the Star algorithm adapted to the various metrics and show that one of the proposed metrics outperforms all others in both effectiveness and efficiency of clustering. We empirically study the sensitivity of the metrics to the threshold value of the algorithm and show improvement with respect to this aspect too.