Most approaches to knowledge discovery consider extractions one by one. However, it is well known that, in practice, any extraction process is iterative. In this work, we introduce an approach that can take benefit from the previously computed extractions in the computation of the current extraction. To do so, we assume that the answers to previous extractions are stored and we use properties of query containment for pruning candidate queries.
In our approach, mining queries are expressed using the notion of context. A context is a triple consisting of two queries and of a set of atomic selection conditions. The two queries define (a) the table according to which supports are computed and (b) the table in which frequent queries are mined, whereas the selection conditions can be seen as constraints on the frequent queries to be mined. Then, we extend the standard operators of the relational algebra to contexts, thus allowing to combine contexts. We show that the computation of composed mining queries can be optimized based on the answers to the mining queries occurring in the composition. We report experimental results which show a significant reduction of computation time, when taking into account the iterative aspects based on our approach.