Hello,
I have a social graph and i have extracted clusterings through community_* methods. I have a mapping vertex->id for ground truth data, meaning that I have a knowledge of the existing communities.
I'd like to understand how similar the communities extracted by the algorithm are similar to the mapping I have from my "ground truth".
I have created a VertexClustering through the FromAttribute method, as I have the mapping as an attribute to each vertex.
My current understanding of the 'danon' method for comparing communities, is that it expects consistency between the ids of two communities. I'll give an example.
Suppose I have vertices 1, 2, 3, 4 ,5.
According to a clustering (e.g. spinglass), i have this mapping:
1->0
2->0
3->1
4->2
5->2
In my ground truth I have this mapping:
1->2
2->2
3->0
4->1
4->1
The overlap between the two communities is perfect, only the labelings mismatch. After all, I don't have control on the IDs that will be used by the community detection algorithm to label communities, and force it to be consistent with mine.
Do I understand correctly that the 'danan' algorithm will consider the similarity of these clusterings as "low"? What could I use given my setup?
Thanks!
Claudio
--
Claudio Martella
address@hidden