MinHash Clustering in Google News Personalization

I've begun looking at the Google News Personalization paper that was presented at the 2007 WWW conference. There are two algorithms evaluated for web-scale clustering of users for use in other collaborative filtering systems: MinHash and Probabilistic Latent Semantic Indexing.

You can find the program notes here: http://www2007.org/paper570.php
and (linked to from the program notes) a PDF version of their paper here: http://www2007.org/papers/paper570.pdf

I find MinHash counterintuitive and confusing, but I'm looking at it more closely now. Some of the resources I've found so far are:

• The Google News Personalization paper above gives a decent introduction to how MinHash works.
• Applying High-Dimensional Clustering Methods For Phylogenetic Profiling mentions using MinHash in the context of clustering genetic pools together to infer species relationships.
• Min-Wise Independent Permutations seems to be a very important paper on the topic.

I'll try to look through all these papers and give a summary of what MinHash is, and how suitable it might be for feature spaces of very large dimensionality, which is a problem space that I happen to be working in at present.