Link prediction in a user–object network based on time-weighted resource allocation

Human dynamics has attracted much attention in recent years. Quantitative understanding of the statistical mechanics of human behavior in an online network is a new challenge for researchers. In an online network, users’ behaviors can be abstracted and projected into a user–object network. Many complex problems concerning resource diffusion, such as recommendation system, network flow and social network behavior, can be solved partially by this user–object network. Although some researchers have given some statistical description of the network recently, little work has been done on link prediction in a user–object network. The objective of this paper is to predict new links based on historical ones in a user–object network. When link weight is taken into consideration, we find that both time attenuation and diversion delay play key roles in link prediction in an user–object network. We then combine these two time effect factors of link weight with users’ lifespans and construct the time-weighted network (TWN) model on the basis of resource allocation. Experimental results show the TWN model can greatly enhance the link prediction accuracy.