A similarity relationships between two objects (so far, either an agent, a signal or a genre, but this could grow). This relationship is pretty general and doesn't make any assumptions on how the similarity claim was derived. Such similarity statements can come from a range of different sources (Musicbrainz similarities between artists, or coming from some automatic content analysis). However, the origin of such statements should be kept using a named graph approach - and ultimately, the documents providing such statements should attach some metadata to themselves (confidence of the claim, etc.).