Using Rough Set Similarity for Link Prediction in Directed Network
Ju Zhang · Systems Engineering · 2015
Given the link prediction in directed network,not only the effect of the common neighbors of two nodes but also that of other nodes in the local community should be taken into account.Firstly,the in-set of one node is defined as the tail-node of directed edges starting with it and the out-set of the node is composed by the head-node of directed edges ending with it.So the local community of a directed edge includes the out-set of its head-node and the in-set of its tail-node.Then we present a rough-set similarity index to measure the possibility of a directed edge for link prediction.After the classification of in-set and out-set with the knowledge of out-degree and in-degree,the rough-set similarity can be determined by the proportion of the difference between their upper approximation sets and their lower approximation sets.Finally using two actual examples of microblogging networks we verify the effectiveness of proposed similarity index,by a comparison with four classic indexes for the precision in the condition of complete links and for the predictability in the condition of missing links.