Detecting Hidden Structure In Social Networks

This 2012 article focuses on the detection of hidden relationship structures in online social networks including the characteristics of relationship structures on Facebook. The authors argue that graphs fail to capture the finer qualities of online social network relationships. Their work is based on the assumption that, of the hundreds of connections, there are only a few important relationships among the many edges within a social network. The authors consider the problem of predicting highest-weight edges occurring in social networks, where the weights measure relationship strength using only inherent topological features of the graph. The authors apply topological metrics from the information retrieval and link prediction literature to rank edges, before comparing the rankings obtained to the true rankings by weight. 

This article will be of particular use to practitioners and researchers interested in narrowing down the important relationships in online social networks such as Facebook. For instance, a Facebook user may have several hundred friends, but in real life only interact with a few dozen people. The authors main tools are algorithms drawn from the literature on link prediction and information retrieval, which estimate the similarity of nodes based on their connected neighbourhoods.