Estimating degree distributions of large networks using non-backtracking random walk with non-uniform jump

This work presents a hybrid sampling method that mixes a non-backtracking random walk and a variation of random walk with jump. We show that the proposed method combines the strengths of both random walks. In particular, the walker of our method will not backtrack to the previously visited vertex so it is likely to produce less number of duplicate samples than the simple random walk. Moreover, the walker's ability to jump ensures that it will explore a network faster. We applied our method on six real world online networks where some of the networks contain millions of vertices. The experimental results show that our method outperformed a non-backtracking random walk and a random walk with jump on estimating degree distributions.