Learning to Assign Credit in Reinforcement Learning by Incorporating Abstract Relations