Data sets in the form of networks occur in many domains such as sociology, biology, engineering, etc. One of the tasks that can be performed on them is the prediction of links, both new as well as recurring. The problem can be tackled by using the graph structure alone or with a combination of node and edge attributes. The attributes which are utilized are mainly domain specific properties of networks. Surprisingly, prior work pays little attention to temporal information.
We investigate the value of incorporating into link prediction methods the history information available on the interactions (or links) of the current social network state.
Our results unequivocally show that time stamps of past interactions significantly improve the prediction accuracy of new and recurrent links compared to rather sophisticated methods proposed recently.