This is an experiment that everyone who learned about cosine distance and vector space information retrieval probably tried, just to see what would happen.

"Fuzzy string matching" isn't really the right term. Of course string distance will give a better fuzzy string match. But this finds words with similar structure rather than words that are similar in the traditional sense. Perhaps there's no obvious application, but it gives results that are very interesting in and of themselves.

It might serve to partition search space for another more expensive algorithm.