I'm trying to think of a good way to explain latent Dirichlet allocation (LDA) to an audience that knows a decent amount about clustering, but nothing about text analysis.
Is it fair to draw a comparison between LDA and fuzzy c-means clustering (not sure if that terminology is official but that's how I've learned it). Are there key differences I should point out in how clusters of text are created as opposed to clusters of other variables?
C-means, like k-means, is really designed for low-dimensional dense data. So I don't think the comparison is fair.
Yet, of course, topic modeling bears a strong resemblance to soft clustering (which is not too popular, users really prefer hard clusterings). But you could use, e.g., these lecture notes which mention that clustering is usually more concerned with each individual point, while topic modeling is more about the topics (word distribution of the entire cluster).
