Asymmetry of social interactions and its role in link predictability: the case of coauthorship networks

The paper provides important insights into understanding the factors that influence tie strength in social networks. Using local network measures that take into account asymmetry of social interactions we show that the observed tie strength is a kind of compromise, which depends on the relative strength of the tie as seen from its both ends. This statement is supported by the Granovetter-like, strongly positive weight-topology correlations, in the form of a power-law relationship between the asymmetric tie strength and asymmetric neighbourhood overlap, observed in three different real coauthorship networks and in a synthetic model of scientific collaboration. This observation is juxtaposed against the current misconception that coauthorship networks, being the proxy of scientific collaboration networks, contradict the Granovetter's strength of weak ties hypothesis, and the reasons for this misconception are explained. Finally, by testing various link similarity scores, it is shown that taking into account the asymmetry of social ties can remarkably increase the efficiency of link prediction methods. The perspective outlined also allows us to comment on the surprisingly high performance of the resource allocation index -- one of the most recognizable and effective local similarity scores -- which can be rationalized by the strong triadic closure property, assuming that the property takes into account the asymmetry of social ties.