When evaluating clustering algorithms for networks, we have well-established metrics like Modularity and Surprise for evaluating the quality of the resulting partition.

If we then embed our graph (well) in a metric space, it becomes much easier to compute distances and communities using geometric algorithms on the points, at the cost of losing our link structure. Therefore, given a partitioning of the points in the metric space (into, say, k clusters) are there any well-established metrics for computing the quality of the clusters?

And if such metrics exist, can we say anything about what embedding/clustering methods are likely to yield clusters that yield similar results when moving from a graph to its embedding?



Your Answer

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