I have a dataset of Facebook users and a set of different clustering algorithms. The project goal is to draw up a rank between these algorithms in order to understand which of them are the good ones. The only information I have is the users dataset and nothing else: no ground-truth, no class / truth labels or whatever. Moreover, each algorithm is designed to create an overlapping clustering, i.e., a given user can belong to more than one cluster.

Since the ground truth is not available, I must use internal indices to evaluate the clustering quality. According to this post: Evaluation measure of clustering (without having truth labels), we can find a lot of indices of this kind in the literature (see this paper for more details). The problem is: all these indices are not useful when the algorithms return an overlapping clustering.

Do you know some indices to evaluate the quality of overlapping clustering?

  • $\begingroup$ For the sake of clarity, what you are saying is not that the clusters that come from different algorithms overlap (which is to be expected), but that you are using algorithms designed to produce overlapping clusters (eg clumping). Is that right? $\endgroup$ – gung - Reinstate Monica May 3 '14 at 21:31
  • $\begingroup$ yes, the algorithms may produce an overlapping clustering result $\endgroup$ – enneppi May 3 '14 at 23:34

Have a look at:

Vendramin, L., Campello, R. J., & Hruschka, E. R. (2010).
Relative clustering validity criteria: A comparative overview.
Statistical Analysis and Data Mining, 3(4), 209-235.

and follow-up work by the same authors. I believe I saw a table there that conained a "feature table" for the various indexes in literature, in particular whether they support overlapping clusters. But I could not find the table (and exact citation) right now.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.