A comparison of graphs of concept for reverse engineering

To group related things together (for example to form subsystems), researchers in reverse engineering are looking for algorithms that create meaningful groups. One such algorithm, concept analysis, received a lot of interest recently. It creates a lattice of concepts which have some advantages over the more traditional tree of clusters from clustering algorithms. We argue that the main interest of concept analysis lies in the concepts themselves and can be disconnected from the particular structure (the lattice of concepts) in which the concepts are usually arranged. We compare concept analysis to various other algorithms trying to select the most important concepts contained in a set of entities. Our main conclusion is that although it has advantages, the lattice of concepts suffers from a major drawback that other constructs do not have: it returns much more information (concepts) than it was given as input (a set of entities describing some software system)