Gradient-based boosting for statistical relational learning: The relational dependency network case