Empirical Analysis of Aging Effects on Preferential Attachment with a Massive Twitter Dataset

Social networks are prominently studied for ad- dressing many pressing challenges, including the spread of diseases and misinformation, orchestrated influence campaigns, and evolution of biological species. A striking common feature of these diverse real networks is the growth of topological structure that is widely understood by the preferential attachment property of scale-free networks that conform to the power law. However, preferential attachment by itself cannot fully explain the phenomenon of aging observed in many complex networks. Therefore, in this paper, we empirically analyzed a massive Twitter dataset and proved that the popularity of nodes tends to "age" over time and that recent nodes get a better chance of forming new connections. We further corroborate these findings with a simple yet effective optimization technique that generates a more accurate in-degree distribution of the Twitter content network. Results show that our approach optimizes the degree distribution of preferential attachment, reducing the RH distance measurement below 0.5 for more than 50% of the Twitter topics in the dataset. By quantifying as an aging coefficient, we also demonstrate that the aging effect is nonuniform across the network during the same time period.