Abstract: This work presents a new perspective on characterizing the similarity between elements of a database or, more generally, nodes of a weighted, undirected graph. It is based on a Markov-chain ...