The Advantages of Seed Examples in First-Order Multi-class Subgroup Discovery