ROCK implementation error

[JosephChataignon]

There is an error in the implementation of ROCK algorithm. In the original paper describing the algorithm, the authors stress the fact that they use the number of links (defined as common neighbours) between clusters, rather than the number of neighbours, when they chose which clusters should be merged. They say one can obtain a matrix of links by multiplying the adjacency matrix by itself. In the current implementation, the goodness measure uses the adjacency matrix where it should use the links matrix (= adjacency_matrix ^ 2).

I am planning to submit a pull request soon to correct this.