Is it possible to combine attribute-value clustering and relational mapping?

Say I want to map an example about vegetables:

1) with att-val descriptors
Tomato has color=red, family=Solanaceae
Carrot has color=orange, family=Apiaceae

2) relations
Node Tomato is related to node Soup, with label "used-for"

I would like to manage the mapping based on this double dimension.
Is it possible in Graphviz, and in what form?


The Graphviz model supports

The Graphviz model supports graphs with nodes and edges, subgraphs of nodes and visual attributes. How you use these depends on your application and its conventions, if any. For abstract (non-realized graphs), there are few constraints, but in making a drawing, there are more restrictions. For example, clustered subgraphs must form a hierarchy.

In general, if you some important general binary relation between nodes, use attributed edges for that. If there is some notion of classes and subset containment, you can use nested clusters. Other attributes can be mapped to shapes, colors and styles.

For ideas, you may want to peruse how various applications in encode their information.

To clarify: the mapping

To clarify: the mapping algorithm should take charge of making groups of most-similar nodes, automatically. Tomatoes should automatically be grouped with peppers, away from carrots, because they share more features.

(i.e. it's not about my  my defining subgraphs manually in the .dof file)


There are data mining viz algorithms for this

but i want a 2-in-1 tool: feature-based automatic clustering + relational graphing.



Graphviz is mostly focused on

Graphviz is mostly focused on relational graphing. Clustering is a vast field on its own. The closest we offer is gvmap, which takes a graph layout and then apples modularity clustering based on the layout. The input graph can specify its own clustering, but that would get you back to providing your clustering algorithm.

Recent comments