I am trying to cluster the graph using spectral clustering. However I am unaware of the number of classes that exist in the data. Will it be a good idea to do PCA on the adjacency matrix to find number of actual clusters in the dataset? Are there any other options?