Software product line engineering is a reuse-driven paradigm for developing families of similar products from a generic product backbone with identified options. A customised product is then derived by combining the artefacts implementing the backbone with the ones implementing the chosen options. Variability analysis and representation is a central task of this paradigm: it consists in suitably defining and structuring the scope, the commonalities, and the differences between the derivable products. Several formalisms have been proposed: some are textual, such as propositional logic or constraint programming, while others are based on annotated graph representations. In this paper, we aim to survey and compare existing graph-based variability representations. Among them, conceptual structures have been used rather early and occasionally employed: this survey highlights their original position, which is due to some of their properties, including canonicity and dual view, that they provide on product configurations versus their features.