Published: 1996 Juli
Buchtitel: Proceedings of the MLnet Familiarization Workshop on Data Mining with Inductive Logic Programming (ILP for KDD) held in conjunction with the 13th International Conference on Machine Learning (ICML '96), Bari, Italy, July 2, 1996
The paper presents an approach for using a bidirectional search strategy for inductively learning clauses in a restricted first-order language. The learning target is to find a set of goal clauses that describe the true ground facts of a given target predicate. In our example setting we further assume that the background knowledge is also given in the form of true (and false) ground facts of each background predicate. By fixing the number of variables allowed in the derived clauses we show that no explicit negative goal facts are needed in the case of the closed-world assumption since the rules are evaluated from the premise to the head rather than binding the variables of the goal literal first. As a consequence we get an efficient ILP algorithm that tries to minimize the tuples of variable substitutions stored at each step of our covering approach.