No problem, you could not guess.
I did what I suggested earlier, that is to say find all diagrams on which the classes on each side of the connector appear, then parse the diagramlinks and hide the one representing my connector. This is not an elegant or optimal solution, but at least it works.
Fo Qwerty, I understand your point of view, but I think my case is an exception. I have a method on how to model use cases, in particular restricting on which diagram(s) each information can appear. However, I want to create some more diagrams for communication purposes, and on those diagrams I want to show some relations that are only implicit with my method (for example, that an element is a sub-element of another). I only create new connectors so that the conceptual link between elements is displayed on the diagram, but they do not add any information, that is why they are hidden except on the diagram for which they were created.