Multi-target Core Network-Based Networked Multi-label Classification

As the increasing popularity of label classification, networked multi-label classification is becoming a hot topic in the field of data mining, where the networked multi-label means that each entity has more than one label during classification in network environments. In the existing works on networked multi-label classification, although only the labels of certain nodes are required to be determined, the labels of all nodes in the network have to be inferred. This works well for small networks, but not for large networks, especially not for large-scale networks with big data, as a plenty of time has been spent to compute a lot of unrequired labels. In this paper, we introduce a core network which is composed of the shortest paths that link some sources (i.e., some nodes with known labels) and some targets (i.e., some nodes with unknown labels required to be determined), as these paths have the most significant directly influence on label classification. Then we propose a novel heuristic MultI-TargeT corE Network discovery algorithm MITTEN to discover a core network, which aims to achieve the relatively accuracy of predicted labels with a relatively short time. Compared with existing networked multi-label classification approaches, the experimental results executed on real networks show that our proposed MITTEN can predict labels in network environments more precisely and more efficiently.