Detecting Strong Cliques in Co-authorship Networks

The study of complete sub-graphs belongs to the classical problems of graph theory. Thanks to sociology, the term clique has come to be used for structures representing a small group of people or other entities who share common characteristics and know each other. Clique detection algorithms can be applied in all domains where networks are used to describe relationships among entities. That is not only in social, information, or communication networks but also in biology, chemistry, medicine, etc. In large-scale, e.g., social networks, cliques can have hundreds or more nodes. On the other hand, e.g., in co-authorship networks representing publishing activities of groups of authors, cliques contain, at most, low dozens of nodes. Our paper describes experiments on detecting strong cliques in two weighted co-authorship networks. These experiments are motivated by the assumption that not every clique detected by traditional algorithms truly satisfies the sociological assumption above. Informally speaking, the approach presented in this paper assumes that each pair of clique nodes must be closer to each other and other clique nodes than to non-clique nodes. Using experiments with weighted co-authorship networks, we show how clique detection results differ from the traditional approach when both the strength of the edge (weight) and the structural neighborhood of the clique are considered simultaneously in the analysis.