Semantic Unlink Prediction in Evolving Social Networks through Probabilistic Description Logic

Recently, prediction of new links between two individuals in social networks has gained a lot of attention. However, to fully understand and predict how the network evolves through time, ending relationships also need to be predicted. Although most approaches use graph-based methods for link prediction, these may not be suited for the unlink prediction task. In this paper, we propose an approach for unlink prediction that uses information about the domain of discourse through a probabilistic ontology, specified in the probabilistic description logic CRALC. We empirically evaluated our approach comparing it with standard graph-based and some state of the art unlink methods. The results shows significant improvement on detecting unlinks when considering our proposal.