Clustering of Paths in Complex Networks

While network analysis is more than 70 years old, the analysis of paths in complex networks is yet almost negligible. Here, we introduce different measures of computing the pairwise similarity of paths, either simply based on the elements in the paths, their sequence, on the graph in which they are embedded, or incorporating all three features. Based on ground-truth in a data set concerning how people solve a one-player puzzle, we show that the classification of the paths using the similarity measures in a hierarchical clustering approach performs best for the similarity measures which integrate all three features. We thus give first evidence that path similarity measures provide another dimension to mine and analyze complex networks.