Weighted undirected graphs, complex Laplacian, complex eigenvalues & spectral clusering

Question

I am rather puzzled and confused, I have been trying to get a clear understanding of how would spectral clustering work for an undirected weighted graph, I have used the normalized Laplacian, but I always get complex not strictly positive eigenvalues, all the resources I am finding build on the results that the Laplacian is real symmetric positive semi-definite matrix, hence real non-negative eigenvalues. Any guidance is greatly appreciated, specially in the answer to the question if I take the norm of the normalized Laplacian would spectral clustering algorithms be still valid with same results.