110 likes | 313 Views
Top Down FP-Growth for Association Rule Mining. Ke Wang Liu Tang Jiawei Han Junqiang Liu Simon Fraser University. Introduction. Association rule A B : A and B: sets of items support: count (AB ) (# of transaction containing AB) Frequent: >= minimum support
E N D
Top Down FP-Growth for Association Rule Mining Ke Wang Liu Tang Jiawei Han Junqiang Liu Simon Fraser University
Introduction • Association rule A B : • A and B: sets of items • support: count(AB ) (# of transaction containing AB) • Frequent: >= minimum support • confidence: count( AB ) / count(A ) • Confident:>= minimum confidence • Input: a set of transactions • find all frequent patterns AB and A
TD-FP-Growth for frequent pattern mining • Similar prefix tree as FP-tree • Items in transactions are sorted • Transactions share prefix as much as possible • FP-growth: bottom-up mining • TD-FP-Growth : top-down mining
root b: 2 a: 3 e: 1 c: 1 b: 1 c: 1 e: 1 c: 1 (b: 1) (b: 1, c: 1) (a: 1, b: 1, c: 1) e: 1 e’s conditional pattern base FP-Growth: Bottom-up minig b, e a, b, c, e b, c, e a, c, d a minsup = 2 Mining order: e, c, b, a
root (b: 1) (b: 1, c: 1) (a: 1, b: 1, c: 1) b: 3 c: 2 • e’s conditional FP-tree must be created separately because counts at upper levels are modified. FP-Growth: Bottom-up mining drawback!
FP-Growth: Top-down mining(TD-FP-Growth) • process nodes at upper level first • counts modified at upper level are not used at lower level • reuse the paths in the original FP-tree for conditional pattern FP-trees See example
root b: 2 a: 3 e: 1 c: 1 b: 1 c: 1 e: 1 c: 1 e: 1 b, e a, b, c, e b, c, e a, c, d a minsup = 2 TD-FP-Growth Mining order: a, b, c, e CT-tree and header table H
b: 1 sub-header-table H_c root a: 2 b: 2 a: 3 e: 1 c: 1 b: 1 c: 1 e: 1 c: 1 CT-tree and header table H e: 1 b, e a, b, c, e b, c, e a, c, d a minsup = 2 TD-FP-Growth
Performance • Data sets from UC_Irvine Machine Learning Database Repository: http://www.ics.uci.edu/~mlearn/MLRepository.html.