ASSOCIATION RULE MINING BASED ON TOPOLOGY FOR ATTRIBUTES OF MULTI-VALUED INFORMATION SYSTEMS

摘要

Association rules mining aims to extract associations and causal structures among sets of frequent items or attributes in a large database. In practice, interesting association rules satisfy predefined minimum support and minimum confidence thresholds. In this paper, we propose a new method to generate association rules which is focused on not only minimum support and minimum confidence thresholds but the shortest length among templates as well. The method is started by a transformation of a multi-valued information system into a two-valued information system. Then, we obtain a binary relation on attributes of the two-valued information system and deduce a topology for the attributes based on the binary relation. Formally, we present two kinds of lattice of the topology for the attributes, i.e., the lattice of the topology and the quotient lattice of the topology which is deduced by the support of subset of attributes. Finally, a new association rules mining method is proposed in the quotient lattice of the topology. Compared with existing association rules mining methods, three contributions of our method were achieved as: (1) all templates of association rules are embedded in the quotient lattice of the topology for attributes; (2) templates with minimum support are shown in the quotient lattice, and association rules with confidence 1 can be mined from equivalent classes of the quotient lattice; (3) association rules with minimum support, confidence 1 and the shortest length among templates can be extracted from the quotient lattice. Examples show that our method is an alternative approach for association rules mining.