The process for detecting hierarchical communities in the skills graph using bootstrapping is described in Figure \ref{830004} in Appendix 2.

Results

As illustrated in Figure \ref{356438}, using the Louvain algorithm with bootstrapping, we generate consensus cluster assignments at three layers. It’s possible to explore finer layers of the hierarchy. However, for the convenience of profiling clusters we stop at depth three where we find 143 clusters. The cluster labels are assigned manually after reviewing the most prominent skills as well as the most common job titles and SOC codes. In future work, we will aim to develop an automatic way of labelling the clusters at all levels.