Just off the top of my head, maybe it could be used for detecting and
applying local optimizations to digital circuits, sort of like
peep-hole optimization in compilers but for hardware. If the circuit
is represented as a graph, then a tool like this could find clusters of
locally connected components that could be replaced by something
simpler. The brute force approach would depend on solving subgraph
isomorphism, which is NP-hard, but the authors claim to have found an
easier way. I've only skimmed the paper so I'm in no position to
judge.
watch the video. it gives examples referring social networks.
User activity monitoring in an event driven architecture. Analytics. Marketing.
Lots of bioinformatics use cases too, getting frequencies of patterns in protein-protein interaction networks for example.