GAP - rule discovery for graded classification

Graded classification occurs in many areas of IT, e.g. user preference, relevance, rating, investment, business competitiveness but also in IR, semantic web, multimedia databases. In this paper we define the graded classification ILP task. Namely having a classified learning data, we want to learn the classification function depending on other attributes. We introduce a procedure based on multiple use of classical ILP with additional qualitative constraints gluing the aggregation function for generalized annotated rules. We consider several preprocessing methods, learning qualitative constraints and provide experiments on benchmark data with different tolerance.