A bipartite graph is a triplet $G=(\top, \bot, E)$ where $\top$ is the set of top nodes and $\bot$ is the set of bottom nodes, and $E\subseteq\top\times\bot$ is the set of edges. Often real-world networks are modelled by a bipartite graph with examples including author/publication graphs, and actor/movie graphs.
In the above examples one generally finds that actors and authors are designated bottom nodes, with publications and movies designated top nodes.
Results such as those found in multiple papers by Matthieu Latapy, and Clémence Magnien, for instance, are then dependent on this convention. By this I mean the following. It is observed that the degree distribution of the bottom set follows a power law, whereas that of the top set is closer to Poisson. One can draw conclusions from this. Perhaps it is reasonable to expect more preferential attachment among live agents. But, it's not difficult to imagine an example on a real-world network which doesn't involve live agents.
There is an obvious symmetry in most real-world networks and it seems that the choice in those publications mentioned, between top and bottom designations tends to favour a neater summary of the results as opposed to having a theoretical basis.
Does such a theoretical basis exist?