The strength of strong ties in scientific collaboration networks

Network topology and its relationship to tie strengths may hinder or enhance the spreading of information in social networks. We study the correlations between tie strengths and topology in networks of scientific collaboration, and show that these are very different from ordinary social networks. For the latter, it has earlier been shown that strong ties are associated with dense network neighborhoods, while weaker ties act as bridges between these. Because of this, weak links act as bottlenecks for the diffusion of information. We show that on the contrary, in co-authorship networks dense local neighborhoods mainly consist of weak links, whereas strong links are more important for overall connectivity. The important role of strong links is further highlighted in simulations of information spreading, where their topological position is seen to speed up spreading dynamics. Thus, in contrast to ordinary social networks, weight-topology correlations enhance the flow of information across scientific collaboration networks. Introduction. – One of the key insights of network theory is that the structure of networks reflects their function and it also sets constraints on dynamical processes taking place on networks [1]. Such structure may be a direct consequence of evolutionary forces acting on the entire system [2,3], such as for modules performing specific tasks in networks of metabolism or genetic regulation [4]. Alternatively, the structure may arise in an emergent fashion from the actions of the individual nodes of the network. This is the case for social networks, where individuals attempt to satisfy their basic social needs related to emotional support, social cohesion, and access to resources and information, while under spatial, time and cognitive constraints [5–9]. In addition, the evolution of networks of social interaction may be influenced by external driving forces; this is especially true for professional networks such as the networks of scientific collaboration considered in this Letter. Social networks are in general characterized by the existence of dense, cohesive social groups that arise out of the above-mentioned individual-level mechanisms and constraints. A prominent mechanism giving rise to dense social groups is triadic closure [10, 11] – learning to know people through the people we know. Simultaneously, the (a)E-mail: [email protected] (b)E-mail: [email protected] interplay of several factors, such as homophily, where individuals of similar characteristics prefer to form ties [10], the need for emotional support and social cohesion, and the high maintenance costs of strong ties give rise to correlations between tie strengths and group structure. The existence of such correlations was hypothesized by Granovetter [6] already in the 1970’s: strong ties are associated with dense network neighborhoods, whereas weak links act as bridges between these. This weak-link hypothesis has since been confirmed with the help of electronic communication records [12–14]. This particular relationship between tie strengths and network structure has several important consequences: first, for the connectivity of the entire network, weak links play a crucial role [15]. Second, because of this, they also act as bottlenecks for diffusion and spreading of information on the network. When compared with a null model where tie strengths are replaced by the network average, simulated spreading of information is slower [12]. However, in networks of professional collaboration, such bottlenecks for information diffusion would act against the purposes of individuals in the network. Whereas networks of scientific collaboration display many characteristic features of ordinary social networks, such as prominent community structure (see, e.g., [16–19]), they are also shaped by different driving mechanisms. First, one can argue that the structure of the underlying space of