Can you give some examples/context please?

By representation do you mean, something reprising a role of a primary subject or merely just an assumption?

It sounds like you are describing the problem of clustering; for any collection of elements that are pairwise related by some measure of similarity, how many clusters are there, and where are the boundaries?

One potential criticism of this formulation, is that it assumes the notion of an element as an axiom from which the cluster-membership of elements is inferred, as well as an a priori metric of similarity. It is a logically atomic model that models clustering as a type of induction.

Perhaps this could be circumvented by treating elements as being another type of relation that is itself a relation between similarity relations(!), such that every subset of relations is understood to be a sub-configuration of a semantically inseparable whole structure. I'm thinking of Tensor Algebra here, as is used in Quantum Field Theory to describe entanglement between an undetermined number of particles that are themselves defined in terms of a global field fluctuation.

It might help if you explain what led you to this question. (Even if the explanation is simply citing a journal article or something like that.)