Sliding-Window Filtering with Constraints of Compactness and Recency in Incremental Database

Abstract

In true-life the database is changed continually in many applications. Incremental mining technique has been developed to avoid rescanning database for knowledge discovery. Recent and compact constraints also are developed for frequent patterns mining. We store the database with a time-vertical bitmap representation, therefore the supports of frequent pattern and recent pattern can be computed fast. Link and bitmap are adopted, so a mass of running time can be saved during incremental mining process. Besides, to mine more efficiently in the incremental database, two concepts of recency and compactness are introduced into sliding-window filtering (denoted as SWF). In essence, an incremental database is divided into several partitions, and a filtering threshold is employed in each partition to handle candidate itemsets generation under constraints of recency and compactness. By employing SWF with constraints of compactness and recency, user satisfactory CFR-patterns (compactness, frequency and recency) can be discovered. Experimental result shows that the running time can be reduced.