Sign prediction based on node connecting tightness in complex network

Nowadays, there are positive and negative relationships among individuals in social networks, which can be abstracted as a signed network with two edge attributes. These two different attributes between individuals are usually used to infer the underlying attitudes of others and modeled as the edge signs in the complex network. Recently, many studies apply local path information to predict edge signs, ignoring the effect of the clustering coefficient of nodes in the three-hop path when determining signs of links. We believe that clustering coefficients of nodes on paths with different lengths reflect the node connecting tightness and contribute significantly to predicting potential relationships. Therefore, we propose a sign prediction algorithm based on the node Connecting Tightness (CT). First, influences of the connection tightness of first- and second-order common neighbors, i.e. nodes on two- and three-hop paths between target nodes, are calculated. Second, based on these influences, the similarity of target node pairs is modeled, and the link sign is predicted by combining the structural balance theory. Finally, node degrees are used to predict directly when there are no common neighbors between the target nodes. Experiments in real-world and artificial networks demonstrate that CT achieves better accuracy and F1 performances in sign prediction.