Supervised temporal link prediction in large-scale real-world networks

Abstract Link prediction is a well-studied technique for inferring the missing edges between two nodes in some static representation of network. In modern day social networks, timestamps associated with each link can be used to predict future links so-far unconnected nodes. these so-called temporal we speak . This paper presents systematic investigation supervised on 26 temporal, structurally diverse, real-world networks ranging from thousands million and links. We analyse relation global structural properties network obtained performance, employing set well-established topological features commonly literature. report four contributions. First, using information, an improvement performance observed. Second, our experiments show that degree disassortative perform better than assortative networks. Third, present new approach investigate distinction modelling discrete events persistent relations. Unlike earlier work, utilises information all past way, resulting substantially higher performance. Fourth, influence activity node or edge differs depending considered type. studied appears most important. The findings this demonstrate how effectively improved explicitly taking into account type connectivity modelled by edge. More generally, contribute understanding mechanisms behind evolution