Analysis of Co-Occurrence Patterns in Data through Modular and Clan Decompositions of Gaifman Graphs
Our team has recently demonstrated that clan decompositions of generalized Gaifman graphs provide advantageous hierarchical visualizations of co-occurrence patterns in data. We develop that early intuition by introducing a construction of implication sets, named "clan implications", and then, explaining these clan decompositions as a variant of closure spaces associated to these implications. Further, we discuss some algorithmic issues that support reasonably efficiently our exploratory data analysis software implementing this approach.
READ FULL TEXT