Using second order information to identify word senses

Most words have multiple senses. 

I was looking at the occurrence of 'bee' in the TASA corpus. Of course, there's the insect, but then a couple of sentences popped up discussing spelling bees.

Of course, this isn't a ground breaking observation.

However, I could use the measure of second order relatedness I'm developing on the contexts that represent a single word. I could cluster the contexts together. Some (arbitrary?) threshold of cluster cohesiveness vs intercluster distance as definition for a given word sense. When comparing meanings between words, compare on the basis of the cluster of word senses that match, rather than also including the usages that don't relate to the word sense under scrutiny.