Spectral bi-partite graph partitioning and clustering
Clustering of large advertiser data set. Rows of the matrix corresponds to advertisers, columns to bidded search terms. Bid values are shown by color entries in the matrix:
a) The original data.
b) The same data matrix after row and column permutations based on a hierarchical spectral graph partitioning (clustering) method.
Image on the left clearly shows emerged sub-markets that group together relevant terms and advertisers.
Overture R&D Technical Report 2003: Spectral Clustering of Large Advertiser Datasets