The predictive value of young and old links in a social network

Abstract

Recent studies show that vertex similarity measures are good at predicting link formation over the near term, but are less effective in predicting over the long term. This indicates that, generally, as links age, their degree of influence diminishes. However, few papers have systematically studied this phenomenon. In this paper, we apply a supervised learning approach to study age as a factor for link formation. Experiments on several real-world datasets show that younger links are more informative than older ones in predicting the formation of new links. Since older links become less useful, it might be appropriate to remove them when studying network evolution. Several previously observed network properties and network evolution phenomena, such as "the number of edges grows super-linearly in the number of nodes" and "the diameter is decreasing as the network grows", may need to be reconsidered under a dynamic network model where old, inactive links are removed.

Link prediction in social networks: the state-of-the-art

“…Then the general relation strength from v i and v j can be calculated by RSS metric. Chen et al [102] also applied a supervised learning approach to study link age as a factor for link prediction. Unlike those that suggest a relatively ad-hoc aging model, here they apply logistic regression to quantify the relative importance of old links and young links.…”