A look at interdisciplinarity using bipartite scholar/journal networks

In this paper, we propose new means to quantify journals' interdisciplinarity by exploiting the bipartite relation between scholars and journals where such scholars do publish. Our proposed approach is entirely data-driven (i.e., unsupervised): we just rely on the spectral properties of the bipartite bibliometric network, without requiring any a-priory classification or labeling of scholars or journals. Our approach is based on two subsequent steps. First, the structure of the bipartite graph is used to co-cluster both journals and scholars in a same low-dimensional space. Then, we measure a journal's interdisciplinarity by computing various diversity metrics (Shannon entropy, Simpson diversity, Rao-Stirling index) over the journal's distance with respect to these clusters. The proposed approach is evaluated over a dataset comprising 1258 journals and 2570 scholars in the information and communication technology field.