Efficient Algorithms for Generating Provably Near-Optimal Cluster Descriptors for Explainability