Resolving author name homonymy to improve resolution of structures in co-author networks

Abstract

We investigate how author name homonymy distorts clustered large-scale co-author networks, and present a simple, effective, scalable and generalizable algorithm to ameliorate such distortions. We evaluate the performance of the algorithm to improve the resolution of mesoscopic network structures, that is those meso-level structures of a network resulting from groupings of nodes and their interlinking. To this end, we establish the ground truth for a sample of author names that is statistically representative of different types of nodes in the co-author network, distinguished by their role for the connectivity of the network. We finally observe that this distinction of node roles based on the mesoscopic structure of the network, in combination with a quantification of the commonality of last names, suggests a new approach to assess network distortion by homonymy and to analyze the reduction of distortion in the network after disambiguation, without requiring ground truth sampling.