Learning to Predict Reciprocity and Triadic Closure in Social Networks
I recently read through a research paper, co-authored by Cornell’s own John Hopcroft, titled Learning to Predict Reciprocity and Triadic Closure in Social Networks. The paper was very interesting given its relevance to the class, especially with its emphasis on triadic closures.
In recent lectures and textbook readings, we have discussed the difference of strong and weak ties, and the very important role that weak ties play as local bridges to connect giant components, as Mark Granovetter wrote in his seminal paper “The strength of weak ties”. We have also discussed the triadic closure, that if there are three nodes A, B, and C, if ties (A, B) and (A, C) exist, then the tie (B, C) is also likely to exist.
This paper goes a step beyond, and explores how these properties differ between typical reciprocal (two-sided/two-way) relationships, which we have been the type of relationships and networks we have been studying in class, and parasocial (one-sided/one-way) relationships, which is novel.
While we know from class that reciprocal relationships are usually those between close friends, the paper interestingly explores the most common example of a parasocial relationship: the one between a celebrity and a fan. This is logical, as we realize that celebrities’ follower counts on social media far exceed the amount of people they follow, which would cause a large amount of one-sided relationships in the social network. The latter, unlike the former, is drawn using directed edges to indicate the one-sided nature of the relationship.
The paper goes even further by exploring how parasocial relationships are developed into reciprocal relationships and finally how these further develop into triadic closures. To do so, they pose the situation graphically as a learning problem, which allows them impressively infer reciprocal relationships in a Twitter network with 90% accuracy. Investigating triadic closures in this network, they also find very interesting results; they note that elite users (users who rank as the top 200 most followed users) play a more important role to form a triadic closure, relative to ordinary users (all other users), showing that triadic closures are three times as likely to form when they involve an elite user.
In conclusion, the paper is an interesting read and its implications far-reaching. I recommend it to others who wish to further delve into the mechanics of social networks in the real world.