Medusa is an approach to detect size-k modules of objects that, taken together, appear most significant to another set of objects. It builds on collective matrix factorization to derive different semantics, and it formulates the growing of the modules as a submodular optimization program.
