Optimal and hierarchical clustering of large-scale hybrid networks for scientific mapping

Previous studies have shown that hybrid clustering methods based on textual and citation information outperforms clustering methods that use only one of these components. However, former methods focus on the vector space model. In this paper we apply a hybrid clustering method which is based on the graph model to map the Web of Science database in the mirror of the journals covered by the database. Compared with former hybrid clustering strategies, our method is very fast and even achieves better clustering accuracy. In addition, it detects the number of clusters automatically and provides a top-down hierarchical analysis, which fits in with the practical application. We quantitatively and qualitatively asses the added value of such an integrated analysis and we investigate whether the clustering outcome provides an appropriate representation of the field structure by comparing with a text-only or citation-only clustering and with another hybrid method based on linear combination of distance matrices. Our dataset consists of about 8,000 journals published in the period 2002-2006. The cognitive analysis, including the ranked journals, term annotation and the visualization of cluster structure demonstrates the efficiency of our strategy.