AUTHOR=Dilmaghani Saharnaz , Brust Matthias R. , Piyatumrong Apivadee , Danoy Grégoire , Bouvry Pascal TITLE=Link Definition Ameliorating Community Detection in Collaboration Networks JOURNAL=Frontiers in Big Data VOLUME=2 YEAR=2019 URL=https://www.frontiersin.org/journals/big-data/articles/10.3389/fdata.2019.00022 DOI=10.3389/fdata.2019.00022 ISSN=2624-909X ABSTRACT=

Collaboration networks are defined as a set of individuals who come together and collaborate on particular tasks such as publishing a paper. The analysis of such networks permits to extract knowledge on the structure and patterns of communities. The link definition and network extraction have a high impact on the analysis of collaboration networks. Previous studies model the connectivity in a network considering it as a binomial problem with respect to the existence of a collaboration between individuals. However, such a data consists of a high diversity of features that describe the quality of the interaction such as the contribution amount of each individual. In this paper, we have determined a solution to extract collaboration networks using corresponding features in a dataset. We define collaboration score to quantify the collaboration between collaborators. In order to validate our proposed method, we benefit from a scientific research institute dataset in which researchers are co–authors who are involved in the production of papers, prototypes, and intellectual properties (IP). We evaluated the generated networks, produced through different thresholds of collaboration score, by employing a set of network analysis metrics such as clustering coefficient, network density, and centrality measures. We investigated more the obtained networks using a community detection algorithm to further discuss the impact of our model on community detection. The outcome shows that the quality of resulted communities on the extracted collaboration networks can differ significantly based on the choice of the linkage threshold.