You can also check out this article . From the DVK answer and this link, it sounds like it's normal graphics, but with “attributes” at the nodes and edges. For example, if you tried to create an image processor, you can first build a graph in which each node represents a clearly colored area (for example, a person’s face against a shirt), and then the edges represent visual adjacency (probably the face / neck is adjacent to both shirt, and to the background of the image).
Then you must assign attributes to the elements of the graph. The attribute for each region (node) can be its predominant color, and the attribute for each edge will be how many of their circles the respective regions share.
As another example, you see a color graph with edge weights - a simple example of an attribute graph.
Perhaps I could be more helpful with a bit more information about why you look in ARG, as they are pretty broadly defined.