Mining cross-table association rules based on projections of itemsets

Association rules are generally recognized as a highly valuable type of regularities and various algorithms have been presented for efficiently mining them in large databases. However, the application of most algorithms is so far restricted to cases where information is put together in single table. For databases containing multiple tables, little work has been done. To solve this problem, we introduce the problem of mining cross-table association rules in two tables containing both quantitative, and categorical attributes, and present an algorithm for mining cross-table association rules. The algorithm is based on the projections of itemsets. All large cross-table itemsets can be constructed by their projections directly. It neither needs to join the tables nor creates a universal table or expends any of the tables. The algorithm orients large databases and considers general relationships expressed by semantic correlation attributes between two tables, where the correlation attributes may be keys, foreign keys or other equivalent attributes. An experiment shows that the algorithm is efficient.