Clustering with Domain-Specific Usefulness Scores