We discovered that the more highly transversal skill groups we removed, the more fragmented the skills network became. It appears that some broad skills serve as a glue connecting more niche skill clusters. Because of this and to preserve as much information as possible, we only remove group 1, which has the lowest specialisation and comprises 66 skills. While they represent only 0.6% of all vertices, the connections between these skills account for 11.15% of all edges in the graph.