Fast Algorithms for Mining Association Rules-英文文献.pdf

Fast Algorithms for Mining Asso ciation Rules Rakesh Agrawal Ramakrishnan Srikant  IBM Almaden Research Center 650 Harry Road, San Jose, CA 95120 Abstract tires and auto accessories also get automotive services We consider the problem of discovering asso ciation rules done. Finding all such rules is valuable for cross- b etween items in a large database of sales transactions. marketing and attached mailing applications. Other We present two new algorithms for solving this problem applications include catalog design, add-on sales, that are fundamentally dierent from the known algo- store layout, and customer segmentation based on rithms. Empirical evaluation shows that these algorithms buying patterns. The databases involved in these outp erform the known algorithms by factors ranging from applications are very large. It is imp erative, therefore, three for small problems to more than an order of mag- to have fast algorithms for this task. nitude for large problems. We also show how the b est features of the two prop osed algorithms can b e combined The following is a formal statement of the problem [4]: Let I = fi ; i ; . . . ; i g b e a set of literals, into a hybrid algorithm, called AprioriHybrid. Scale-up 1 2 m exp eriments show that AprioriHybrid scales linearly with called items. Let D b e a set of transactions, where the numb er of transactions. AprioriHybrid also has ex-