Correlation clustering: A parallel approach?

Correlation clustering is a NP-hard problem, and for large graphs finding even just a good approximation of the optimal solution is a hard task. In previous articles we have suggested a contraction method and its divide and conquer variant. In this article we present several improvements of this method (preprocessing, quasi-parallelism, etc.) and prepare it for parallelism. Based on speed tests we show where it helps the concurrent execution, and where it pulls us back.