Learning Relations by Path Finding and Simultaneous Covering

In this paper, we propose a simultaneous covering inductive logic programming approach RPBL (relational paths-based learning). It uses relational paths as explanations for positive training examples to overcome the explosion problem existed in standard inductive logic programming. It benefits greatly from a simultaneous covering strategy, and avoids local maxima and local plateaus. Experimental results show that it not only performs better than standard approaches such as FOIL, but also than some state-of-the-art approaches.