Mining for association rules in large databases is an important data
mining problem. Many algorithms have been proposed since the problem
was first introduced in 1993. Basically, the procedure of mining
association rules can be divided into two steps: first find out all
frequently appearing sets of items (large itemsets), then calculate
association rules using the large itemsets. Once all the large
itemsets are obtained, the generation of association rules is quite
simple.
In Jiang, Inaba, Imai\cite{jiang-inaba-imai}, we first proposed a new
approach for generating large itemsets by using Binary Decision
Diagram (BDD). Extensive experiments using BEM-II\cite{minato}
package are conducted to evaluate large itemset generation
performance, and the results show that BDD represents and manipulates
the large itemsets efficiently. In this paper, we will first review
some of our previous work, then show how to improve the basic BDD
algorithm\cite{jiang-inaba-imai} to cope with the actual large
database. At last, we discuss the possibility of applying BDD to
sampling method. Also the relation between the prime implicants of
BDD and the candidate sets will be carefully examined.