|
Abstract : |
Many knowledge discovery (kdd) systems need to spend substantial amount of effort to search for rules and patterns within large amount of data. After some natural evolutions, as a consequence of updates applied to their databases, these systems must update their previously discovered knowledge to reflect the current state of their databases. The straight forward approach of re-running the discovery process on the whole updated database to re-discover the rules and patterns is not cost-effective in general, and is unacceptable in many cases. We have studied the problem of updating discovered association rules and found that it is nontrivial, because updates may not only invalidate some existing strong association rules but also turn some weak rules into strong ones. An incremental technique and a fast algorithm FUP have been proposed previously for the update of discovered single-level association rules. In this study, a more efficient algorithm FUP*, which generates a smaller number of candidate sets when comparing with FUP, has been proposed. In addition, we have demonstrated that the incremental technique in FUP and FUP * can be generalized to some other kdd systems. An efficient algorithm MLUp has been proposed for this purpose for the updating of discovered multi-level association rules. Our performance study shows that MLUp has a superior performance over the representative mining algorithm such as ML-T2 in updating discovered multi-level association rules. 1, |
