Considering that spectral clustering algorithms are unable to efficiently discover communities in massive networks for the high computation cost, a spectral clustering ensemble algorithm(SCEA) is proposed. Firstly, the effective KASP algorithm is used to produce clusters. Then the better clusters are chosen to construct a cluster ensemble, and the similarity graph for individual cluster is created. Finally, the resultant overlapping communities with hierarchical soft clustering are obtained. Experimental results show that, in contrast with some typical algorithms such as clique percolation method(CPM), Link, community overlap propagation algorithm(COPRA) and sampled spectral distance embedding(SSDE), the SCEA can discover the network communities with higher normal mutual information(NMI), and it exhibits satisfactory robustness.