Sampling in association rule mining
Tsau Young Lin · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2004
A relation is a representation of a set, called the universe V, of entities by a set of tuples. Hence it is associated with a unique sub-lattice, called relation lattice, of the partition lattice of V. In this paper, we examine the relation lattices on V and a sample V'(a subset). The analysis concludes that only very special types of samples may be able to have the "same" association rules as those of the original universe; here "same" means allowing some statistical errors. In other words, finding association rules by mere random sampling may not be able to fully reflect the association rules of the original universe; special attentions on the sampling are needed.