Evidence for widespread thematic structure in the mental lexicon: supplemental information

Many real networks exhibit parts that are more tightly connected than others. The goal of network clustering is to identify clusters with members that are more strongly inter-connected than connected to the rest of the network. The resulting solution should thus exhibit denser interconnectivy within clusters than between clusters. Network clustering is a promising technique that only recently has allowed us to investigate large-scale directed and weighted graphs. At the same time it is notoriously complex and a single metric to determine the structure in a graph is unlikely to be applicable to any kind of graph (see the extensive review of the state-of-the-art in Fortunato, 2010). However, at the moment of writing the Order Statistics Local Optimization Method (OSLOM) leads to superior results on directed graphs and the detection of strongly overlapping clusters, which makes it particularly suited for the study of graphs derived from word association data. Below we provide an informal summary of how this algorithm works, but refer to the original article (Lancichinetti, Radicchi, Ramasco, & Fortunato, 2011) for full details. The key feature of OSLOM in comparison to other approaches is that it uses th