Fast updated frequent-itemset lattice for transaction deletion

摘要

The frequent-itemset lattice (FIL) is an effective structure for mining association rules. However, building an FIL for a modified database requires a lot of time and memory. Currently, there is no approach for updating an FIL with deleted transactions. Therefore, this paper proposes an approach for maintaining FILs for transaction deletion without rescanning the original database if the number of eliminated transactions is smaller than the threshold determined based on the pre-large and diffset concepts. A diffset-based approach is first used for fast building an FIL. Then, two proposed approaches (tidset-based and diffset-based) are used for updating the FIL with transaction deletion. The experiment was conducted to show that the diffset-based approach outperforms the tidset-based and the batch-mode approaches.