Uncovering the Mechanisms of Co-Authorship Network Evolution by Multirelations- Based Link Prediction

Introduction and literature review Co-authorship network, a proxy of research collaboration, reveals the collaboration patterns and the determining factors through social network analysis perspective, with nodes representing authors and links representing co-authorships (Ortega, 2014; Yan & Ding, 2009). If we know what mechanisms push the evolution of coauthorship network, we could predict which authors may collaborate in future. Most of the studies correlate co-authorship evolution mechanisms to similarity indicators which quantitatively compared by link prediction in homogeneous network (Lu & Zhou, 2010). In order to integrate multirelations between authors, pathbased similarity indicators are proposed for coauthorship prediction in DBLP heterogeneous network (Sun et al., 2011; Sun & Han, 2013). However, what is the role of each mechanism plays and how to combine multiple mechanisms to suit the co-authorship network evolution need to be clarified, moreover, the method need to be verified in different domains. Therefore, we integrate similarity indicators based on multirelations in heterogeneous network and quantitatively evaluate them by link prediction justly, to uncover and infer the mechanisms of coauthorship network evolution. Firstly, similarities between authors are represented by a matrix where the rows are multirelations and the columns are multirelations’ measures. Secondly, the evaluation of similarities is processed based on link prediction, to reveal the importance of each mechanism which is the weight for combining multiple mechanisms. Finally, experiments are presented in the domain of Library and Information Science (LIS), which reveals the best appropriate mechanism, the significance of each mechanism and the combination strategy of different mechanisms.