Pairwise normalization in Simrank variants: Problem, solution, and evaluation

Abstract

Despite of the success in the real-world applications, SimRank and its variants, rvs-SimRank and PRank, suffer from the pairwise normalization problem (PNP) as a counter intuitive property hidden in their computation paradigm. JacSim, a state-of-the-art measure, provides an effective solution to PNP; however, it considers PNP and the effectiveness of the provided solution only in SimRank. In this paper, we consider PNP in SimRank variants, generalize the solution to those measures, conduct extensive experiments with two real-world datasets to accurately perform parameter tuning, and verify the effectiveness of applying the solution to SimRank variants with both un-weighted and weighted graphs.