Towards a Finer Assessment of Extraction Contexts Sparseness

It is widely recognized that the performances of frequent closed itemset mining algorithms are closely dependent on the type of handled extraction contexts, i.e., sparse or dense. In this paper, we address an important question: how can we formally define the sparseness of a given extraction context and assess its value? As an answer, this paper presents a study in which we deal with the problem of assessment of an extraction context's sparseness. Indeed, using the framework of the Succinct system of minimal generators, we present a new sparseness measure which results from the aggregation of two complementary measures, namely the succinctness and compactness measures of each equivalence class, induced by the closure operator. Preliminary experiments mainly permit to rectify the classification of benchmark contexts and confirm our claim that the "dense" and "sparse" qualifications are not absolute ones.