It ultimately depends on whether you want information entropy or thermodynamic entropy. Either way, I would suggest taking a graph theory approach, since the structure of a graph IS a network (or one could say that the structure of your network IS a graph!). That said, it would depend on what kind of graph you want to use to model your network, and then it will depend on what kind of entropy measure you want to try to use for your system.
Here's a nice review of entropy measures on graph networks.
Here's a specific example for a random graph using Shannon entropy.
There are several well-defined measures of entropy for graphs, and the literature is abundant with examples. If you specify a bit more, then you might get better help here. (what is a "natural" network?)
Another option within the graph theory approach is to simply use statistical mechanics on the graph (i.e. microcanonical ensemble or canonical ensemble) and derive the thermodynamic entropy from it. This article has some great examples of how to do so.
EDIT: Per your comment, there do exist topological measures of entropy for graphs. It seems that you have some options: either you use 1) the "graph entropy" definition introduced by Korner on your network, or 2) view your graph topologically, and then use the "topological entropy" definition. It seems you desire to use the latter approach.
I would like to know if there are graph theoretic topological measures too.
This review contains lots of great references that you can explore, and on page six they examine topological entropy of graphs by looking at how "graph maps" behave. If this is unfamiliar to you, consider the famous quote, paraphrased here, by Hermann Weyl: "if you want to understand some mathematical structure, look at its group of automorphisms." And this paper is nice to keep the various ideas of entropy separate.