Inducing Multi-Level Association Rules from Multiple Relations
The paper presents a hypothesis language declaration formalism and a corresponding refinement operator that successfully combines the levelwise search principle with a first-order hypothesis language and thus provides an improvement to the so-called optimal refinement operators that are commonly used in descriptive ILP. The refinement operator is based on the candidate generation procedure of the Apriori algorithm. It extends the Apriori candidate generation in that it allows to define constraints on the combinations of literals in the hypotheses. Experimental results show the usefulness of the approach.