•The method, as
published (R.A. Jarvis and E.A. Patrick, Clustering using a similarity method based on shared nearest
Transactions on Computers C-22 (1973)
1025-1034 ) works like this:
–For each item, find its J nearest neighbours. This requires O(N2) CPU time,
but needs to be done only once. The Daylight implementation is closer to O(NlogN) generally.
–Two structures cluster together if: (a) They are in each
other's list of J nearest neighbours,
–and (b) K of their J nearest
neighbours are in common.