Efficient Discovery of Multiple-level Patterns.
Francesca Alessandra Lisi, Donato Malerba · CINECA IRIS Institutional Research Information System (University of Bari Aldo Moro) · 2002
In the context of frequent pattern discovery, the availability of concept hierarchies over objects of interest requires the development of ad-hoc algorithms for dealing with multiple levels of description granularity. We present a novel framework that allows a uni ed approach to both relational and structural features of data. Patterns are intended as unary conjunctive queries and ordered according to the relation of query subsumption. A re nement operator for searching these pattern spaces is de ned and eAEciently implemented in the candidate generation phase of SPADA, a system for mining multiple-level association rules from spatial data. Experimental results show a remarkable improvement of the overall performance of the system.