Link-based methods for bibliometric journal ranking

The ISI Impact Factor is widely accepted as a possible measurement of academic journal quality. However, much debate has recently surrounded this use, and several complex alternatives have been reported. In this paper, a link-based framework for academic output is proposed, where publications, journals and authors are represented as three sets of nodes in a multi-layered and inter-connected network. Several existing indicators of journal impact are analysed and redefined by the links between the nodes. Furthermore, the indicators are combined and transformed to fused-links between journals, which are further applied to supervised and unsupervised learning methods in order to evaluate impact as well as predict ranks of journals. The link-based framework is explicable and intuitive. The experimental evaluation demonstrates that by applying the proposed fused-links to K-means clustering, the detected clusters are consistent with the ranking of expert reviewers. For journal rank prediction, the accuracy of modified K-nearest-neighbour approaches based on the fused-links is greater than other conventional methods such as the decision tree based approach.