Local Similarity and Community Paradigm: The Robust Methods toward Link Prediction

In this work, we propose a method collaborating the local similarity and local community paradigm with a tunable parameter to balance the contribution of the energy from these two sources. We show that local similarity e.g., common neighbors and local community paradigm e.g., local community links both play significant roles in network evolution; therefore, one cannot ignore or penalize anyone of these two. As different networks are evolved according to different preferences, either local similarity or local community links cannot effectively predict the missing links of all networks. By combining these two sources of information with a tunable parameter, we can balance the power of both of them to obtain remarkably robust methods to predict the missing links. The results have been shown to outperform and more robust than any single of these traditional approaches on most of the networks utilized in this article. Moreover, the proposed method can also reveal the energy of community connections such that to what extent they are involving in the emerging of new links.