A Sample-Based Approximate Ranking Method for Large Graphs

Graph centrality can be estimated by random walk algorithms such as PageRank. These algorithms are useful when ranking items with relationships, and are based on global structure of the graph. It needs to traverse the entire edge-set more than once to get a convergent result. However, when it comes to large graphs, traversing the whole graph is costly, especially when a single node's information is wanted. How to get a rough look at the rank of a specific node? Can the methods be simpler that don't bother to scan the entire graph? How to scale up the method to get more information of the graph? In this paper, we proposed a simple and light algorithm to get approximate rank value of a node. It could be simplified by sampling and expanded to distributed environment. Experiments on representative public graphs have shown that the method can provide an approximate ranking. And the higher ranking a node has by global algorithms, the more accurate the approximate rank will be. Our method also reduces the computation to several orders of magnitude than traditional global algorithms. With sampling techniques, it can outperforms the basic method by additionally reducing computation. It is useful in both estimation and edge computing.