A classification of overlapping clustering schemes for hypergraphs

In reality, friend groups often overlap, and some publications do belong both to biology and chemistry. Yet the most commonly used and studied community detection methods assume that all communities are distinct, with binary assignments to communities.

In this paper, we continue a line of research into classifying community detection methods with desirable properties into the area of overlapping partitions. Unlike the optimization-based methods like modularity optimization, which are computationally infeasible to compute exactly and can have very poor behaviour in certain cases, this gives us a very tractable family of methods with guarantees on their behaviour.

An earlier version of this paper was included in my licentiate thesis, and was presented at PCC 2024.