Ecological Patterns in Online Social Interactions

This thesis builds upon the framework developed by Grilli in "Macroecological laws describe variation and diversity in microbial communities", which identifies a set of statistical patterns present in microbial communities. Following the analogy between natural ecosystems and so-called “information ecosystems” proposed by Plata et al. in "Neutral theory for competing attention in social networks", we explore whether similar laws exist in social systems, specifically, in the temporal dynamics of memes and hashtags on social media. Focusing on different data streams from the online platform Twitter/X, we analyzed five core patterns commonly found in macroecological systems: the Mean Abundance Distribution (MAD), Taylor’s Law, the Abundance Fluctuation Distribution (AFD), the Species Abundance Curve (SAC), and the Short-Term Abundance Change (STAC). Our results reinforce the analogy proposed by Plata et al., demonstrating that these patterns also hold for information ecosystems across all datasets considered. To identify the main drivers behind this result, we model the dynamics of different memes using a Stochastic Logistic Model (SLM). The model assumes no interactions between hashtags but allows for heterogeneity among them. Two approaches were used to generate simulations: in the first, the SLM parameters—carrying capacity (K) and the coefficient of variation of growth rate fluctuations (σ)—were inferred directly from each real time-series; in the second, they were estimated by manipulating the empirical MAD. In both cases, the simulated datasets reproduced the same functional forms observed in the original data (e.g., LogNormal-shaped MADs remained LogNormal). To test the robustness of the model, we introduced a perturbation by artificially “drugging” the abundance of a randomly selected hashtag while keeping all others unchanged. This simulates external interventions or atypical events in real social systems, such as paid promotion, coordinated campaigns, or algorithmic amplification. We then analyzed how this targeted anomaly affects the five macroecological patterns to identify which are most sensitive to externally influenced behavior. This approach provides insight into how detectable departures from the expected statistical structure can serve as indicators of non-organic dynamics in complex social systems.