Adaptive degree penalization for link prediction

Many systems of interest are best described using networks that represent binary relationships among their elements. Link prediction aims to infer the link formation process by predicting missed or future relationships based on currently observed connections. Different techniques and measures have been proposed in the literature to solve this problem. Similarity-based local methods achieve high precision with a low computational complexity. However, determining which particular technique should be applied for each particular network remains an open question. In this paper, we exploit the existence of a relationship between the best-performing degree of penalization for shared neighbors and the network clustering coefficient. We propose an adaptive degree penalization link prediction method, a novel link prediction technique that achieves better results than previously proposed methods.