I am trying to identify 6 clusters in a graph of authors. Authors are connected with an edge when they have co-authored on a paper. I have already created a clustering with K-means (6-Means) and with Agglomerative clustering. Now, I want to combine these two clustering approaches in order to get more accurate clusters. I am able to find corresponding clusters, but I have no clue on how to decide to which cluster an author belongs when the algorithms do not agree.

Possibly I could use some form of a linear combination, but I do not know how to apply this properly. Other methods might be possible, but I am not sure on that.

  • $\begingroup$ The bozo solution might be to run one algorithm then create an additional feature "cluster_label" and feed it into the next algorithm. You could then scale the "cluster_label" feature up or down until you get the result you want. Possibly experiment with the ordering of the two. $\endgroup$
    – AN6U5
    Commented Jan 13, 2016 at 19:59
  • 1
    $\begingroup$ en.wikipedia.org/wiki/Consensus_clustering $\endgroup$
    – Emre
    Commented Jan 13, 2016 at 20:14

1 Answer 1


Since you have a graph of co-authors, it might make more sense to frame the problem at spectral clustering which is graph based.

Then, you can apply something like consensus clustering to combine the clusters.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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