Multi-target tracking applied to evolutionary clustering

We extend an established and robust method from multi-target tracking showing how it can be used for evolutionary clustering. Our framework models the real-life dynamics of consumer web data: the number of objects grows with time, and not all objects update their state synchronously. Our proposed algorithm tackles this problem by estimating the clusters sequentially using methods of multi-target tracking. We compare this novel technique to clustering algorithms commonly used in the literature and show how our method outperforms the other methods in terms of accuracy, stability and speed of adaptation to group dynamics. Our algorithm successfully detects changepoints in the number of clusters.