The PageRank model, initially proposed by Google for search engine rankings, provides a useful network centrality measure to identify the most important nodes within large graphs arising in several applications. However, its computation is often very difficult due to the huge sizes of the networks and the unfavourable spectral properties of the associated matrices. We present a novel multi-step low-rank factorization that can be used to reduce the huge memory cost demanded for realistic PageRank calculations. Finally, we present some directions of future research.