Beyond Isomorphism: Mining Frequent Subgraphs up to Graph Class Membership

This paper introduces novel approaches to the frequent subgraph mining problem using graph class membership. We develop two main frameworks. The first mines frequent subgraphs up to graph equivalence, illustrated by the concept of homeomorphism of arc-weighted directed graphs. The second mines frequent subgraphs up to membership in a general chain-hereditary graph class that admits an efficiently computable canonical labelling. Both approaches enable more flexible and scalable mining by leveraging type-based abstraction rather than exact isomorphism.