Formal concept mining: A statistic-based approach for pertinent concept lattice construction

Abstract

In this paper, we define formal concept mining, a method for generating and evaluating all the pertinent concepts from large transaction databases. We propose a novel efficient formal concept mining algorithm, called Distribution Curve Self-Evaluation (DCSEA). Attempting repeatedly to self-adjust the normal distribution curve to be as close as the symmetry curve, DCSEA automatically identifies all the pertinent concepts by deleting and masking non-pertinent concepts. Instead of using the global support threshold, DCSEA allows users to specify the interestingness of the output concepts by using a more understandable statistic-based threshold, called minimum significance threshold. Such threshold measures the level of significance of the concept extent size (the number of objects) from all the concept extent sizes. Experimental results showed that the proposed algorithm gives high concept retrieval performance, and efficient concept focusing, especially on large databases. © Springer-Verlag Berlin Heidelberg 2004.