Collaborative filtering approach to link prediction

Link prediction is a fundamental challenge in network science. Among various methods, local similarity indices are widely used for their high cost-performance. However, the performance less robust: some networks highly competitive to state-of-the-art algorithms while other they very poor. Inspired by techniques developed recommender systems, we propose an enhancement framework based on collaborative filtering (CF). Considering delicate but important difference between personalized recommendation and link prediction, further improved named as self-included (SCF). The SCF significantly improves accuracy robustness of well-known indices. combination simple index can produce with much lower complexity compared elaborately-designed algorithms.