Comparing Network Centrality Measures as Tools for Identifying Key Concepts in Complex Networks: A Case of Wikipedia

Abstract

Network centralities are amongst the most important measures for tracking and locating crucial nodes in a network. In this paper, we propose a general approach for identifying the most suitable centrality measure for detecting key concepts in a semantic or linguistic network. We experiment with seven network centrality measures (degree centrality, betweenness centrality, closeness centrality, eigenvector centrality, current-flow betweenness centrality, current-flow closeness centrality and communicability centrality). For the purpose of evaluation, we compare the original Wikipedia hyperlink network with a constructed concept network. The obtained results indicate that all seven used measures have good potential for identifying key terms, and that degree centrality achieves the best score. A good score is also obtained for current-flow betweenness centrality and current-flow closeness centrality.