1, k-means
most popular clustering algorithm, but the deficiency is the need to specify k.
2, k-medoids
using similarity matrix, find a node in the current cluster leading to the least sum distance.
3, spectral clustering
first do dimension reduction, and then apply k-means.
seems to perform better than k-means.
4, hierarchical clustering
binary tree. from top to down or from down to top. and huge cost
Shirley's Blog 