Reciprocity in Social Networks: Measurements, Predictions, and Implications

Understanding and modeling the mechanisms by which directed social networks evolve are active areas of research. One related emerging topic is to understand and predict the formation of reciprocal edges, which has many potential applications such as directed social network modeling, friend recommendation, information propagation and targeted spamming. However, there have been few studies providing a systematic understanding of reciprocity at scale. In this paper, we bridge this gap using a novel large-scale Google+ dataset crawled by ourselves as well as one publicly available social network dataset. First, we investigate the structure and evolution of reciprocal and unreciprocal edges. Specifically, we find that reciprocal edges mainly connect users with similar degrees while unreciprocal edges mainly link ordinary users (e.g., users with low degrees) and popular users (e.g., celebrities). However, the impacts of reciprocal edges linking ordinary and popular users on the network structures become increasingly significant as the social networks evolve. Second, we observe that factors including user behaviors, node attributes and edge attributes all have significant impacts on the formation of reciprocal edges. Third, in contrast to previous studies that treat reciprocity prediction as either a supervised or a semi-supervised learning problem, we identify that reciprocity prediction is better modeled as an outlier detection problem. Finally, we perform extensive evaluations with the two datasets, and show that our proposal outperforms previous reciprocity prediction approaches.