Using Reduced Itemset Lattice To Support Online Mining For Dynamic Databases

Abstract

Current online techniques of mining association rules have incurred the problems of huge space requirement for the itemset lattice and of limiting the number of items in rules or producing an incomplete set of rules. In this paper, we propose an improved structure, namely RIL (Reduced Itemset Lattice), to keep just enough itemsets to support online mining for dynamic databases. Once built with the original database, the RIL incrementally adapts itself to database updates. Instead of rebuilding RIL, a complete set of association rules can be directly mined, or expanded first if necessary, from the RIL in a short time. Experimental evaluation shows that the RIL can effectively reduce the space of itemset lattice and mine the rules online for dynamic databases.