This document summarizes a seminar presentation about compression-based graph mining that exploits basic structural primitives like triangles and stars. It discusses how social media graphs are very sparse, with only a small fraction of possible edges existing. The technique codes graphs by representing recurring substructures like hubs and meshes compactly. This clustering reveals the graph's transitivity and hubness. Outcomes include seeing which basic structure is most common and getting an overall minimum graph clustered by dense areas. While results seem good, the document notes critics that example codings are not shown and probabilities given are not replicable.