Computing Frequent Itemsets Inside Oracle 10G

Frequent itemset counting is the first step for most association rule algorithms and some classification algorithms. It is the process of counting the number of occurrences of a set of items that happen across many transactions. The goal is to find those items which occur together most often. Expressing this functionality in RDBMS engines is difficult for two reasons. First, it leads to extremely inefficient execution when using existing RDBMS operations since they are not designed to handle this type of workload. Second, it is difficult to express the special output type of itemsets. In Oracle 10G, we introduce a new SQL table function which encapsulates the work of frequent itemset counting. It accepts the input dataset along with some user-configurable information, and it directly produces the frequent itemset results. We present examples of typical computations with frequent itemset counting inside Oracle 10G. We also describe how Oracle dynamically adapts during frequent itemset execution as a result of changes in the nature of the data as well as changes in the available system resources.