Structural communities in Italian Computer Science academia: Which relation with scientists’ socio academic and personal data? What impact on performance?

Until now, the literature on community detection in informetrics has focused its attention mainly on the development of faster and reliable algorithms, and on the comparison between structural communities (detected from co-authorship networks) and topic communities (detected from “author-topic” networks). To the best of our knowledge, only a work by Marko A. Rodriguez and Alberto Pepe (2008) investigates the relation between structural communities, detected from the co-authorship network, and socioacademic communities, identified by the scientists’ academic rank and university. While this study focuses only on the comparison between these two communities, the one that we propose investigates how the membership in a socio-academic community could impact on the membership in a structural community. In this work we analyze the relation between some socio-academic data, and some characteristics of the structural communities such as size, cohesion, and diversity (resulting from the analysis of gender, nationality, and university of each member of a structural community). In addition to the socio-academic data, we evaluate also the impact of other scientists’ personal data, such as gender and age, on the above said characteristics of their structural communities. The analysis of the impact of scientists’ socio-academic and personal data could improve our knowledge about the development of research teams and scientist social capital, whose importance has been growing in the last decades, given the increasingly collaborative nature of science. We are interested also in investigating the impact of the structural communities’ characteristics on the scientists’ performance. Even if the relation between scientists’ social capital and their productivity has been widely discussed in the literature, so far no one has analyzed the impact of the membership in structural community on performance. In order to answer these research questions, we collect personal and socio-academic data of all the population of Italian Computer Science academics (almost 800 assistant, associate and full professors). Then, we detect via Scopus their relevant co-authors in 2006-2010, i.e. the scientists who have co-authored at least three of their publications in this period, with an Italian Computer Science academic. Next, we collect via Scopus the complete list of publications of Italian Computer Science academics and of their relevant co-authors. We then construct an “author-publication” matrix of dimensions m×n, with m higher than 2,300 and n higher than 135,000. We employ different community detection algorithms, choosing among edge, betweenness, fast greedy, leading eigenvector, label propagation, walk trap, and spinglass, to assign a structural community to each scientist in the network. In order to adopt a more reliable indicator, scientists’ performance is measured by Fractional Scientific Strength (Abramo et al., 2013) starting from the 2006-2010 publications indexed in WoS