A First - Order Approach to Unsupervised Learning 3

This paper deals with learning rst-order logic rules from data lacking an explicit clas-siication predicate. Consequently, the learned rules are not restricted to predicate deenitions as in supervised Inductive Logic Programming. First-order logic ooers the ability to deal with struc-tured, multi-relational knowledge. Possible applications include rst-order knowledge discovery, induction of integrity constraints in databases, multiple predicate learning, and learning mixed theories of predicate deenitions and integrity constraints. One of the contributions of our work is a heuristic measure of connrmation, trading oo satisfaction and novelty of the rule. The approach has been implemented in the Tertius system. The system performs an optimal best-rst search, nding the k most connrmed hypotheses. It can be tuned to many diierent domains by setting its parameters, and it can deal either with individual-based representations as in propositional learning or with general logical rules. We describe a number of experiments demonstrating the feasibility and exibility of our approach.