Structural Graph Clustering: Scalable Methods and Applications for Graph Classification and Regression

This thesis focuses on graph clustering. It introduces scalable methods for clustering large databases of small graphs by common scaffolds, i.e., the existence of one sufficiently large subgraph shared by all cluster elements. Further, the thesis studies applications for classification and regression. The experimental results show that it is for the first time possible to cluster millions of graphs within a reasonable time using an accurate scaffold-based similarity measure.