Information Diffusion in Social Networks

Background

Social networks as abstract means of representing relations and communications within and between social groups of arbitrary size, and information dissemination in them, have been an object of active studies for the greater part of the XXth century—see, e.g. Davis (1969) and especially Rogers (1995) who reviewed 506 diffusion studies, though without proper sociometric information. In what follows I present a brief overview of key ideas and key authors. Readers interested in the subject are advised to consult additional literature listed at the end of the chapter.

The issue of information diffusion in small-world social networks was first seriously studied by Mark Granovetter in his seminal paper “The Strength of Weak Ties” (1973) and the follow-up paper “The strength of weak ties: a network theory revisited” (1983). Granovetter suggested that the main information exchange takes place along the so-called weak ties—loose, “acquaintance-style” connections between social network members, while strong (“friendship-style”) ties are responsible for decision-making and knowledge generation and preservation. The theory of weak and strong ties serves as a basis for contemporary information diffusion theories.

Information in social networks spreads in the form of messages, wall posts, and even so-called one-bit pokes researched by Kaye et al. (2005) and Xiao et al. (2007). Golder et al. (2007) published the first massive study of message dynamics in Facebook and discovered that the messages follow regular temporal patterns based on time of day, day of week, and season. The origins and the destinations of the messages are not random, suggesting that the dissemination processes are homophily driven. The findings of Golder et al. (2007) were corroborated by Singla & Richardson (2008) and Leskovec & Horwitz (2008) who explored homophily and planetary-scale message traffic in Microsoft™ Instant Messenger, respectively. Gruhl et al. (2004) observed similar temporal message patterns. In addition, Leskovec & Horwitz have shown that there is a dependency of communication patterns on geographical proximity of the correspondents at the session-level (rather than message-level): longer distances yield fewer and shorter conversations.

Navigability in complex networks —i.e., efficient mechanisms for finding intended communication targets—was modeled by Boguña et al. (2008). The authors proposed a concept of a hidden metric space underlying a complex network, with its own coordinates and distances that serve as guidelines for making routing decisions. At the observable level, navigability translates into scale-free node degree distributions and strong clustering, both properties being common in social networks—which are, thus, navigable.

Kossinets et al. (2008), based on the work of Gibson (2005), noticed that the members of a social network do not continuously participate in communications with their neighbors. Information only spreads as a result of discrete communication events and the frequencies of these events make strong influence on the preferred communication pathways. These pathways form a communication backbone that dynamically changes to reflect the momentary fluctuations of the event frequencies.

Wilson et al. (2009) further elaborated the concept of the communication backbone by correlating social and interaction degrees of network members and discovering that they do not scale uniformly.

Viswanath et al. (2009) traced the mechanisms that trigger communications in online social networks and discovered that 54% of the interactions are attributed to site mechanisms (such as birthday reminders) and that most links in a communication backbone rapidly die out.