Long Trend Dynamics in Social Media

A main characteristic of social media is that its diverse content, copiously generated by both standard outlets and general users, constantly competes for the scarce attention of large audiences. Out of this flood of information some topics manage to get enough attention to become the most popular ones and thus to be prominently displayed as trends. Equally important, some of these trends persist long enough so as to shape part of the social agenda. How this happens is the focus of this paper. By introducing a stochastic dynamical model that takes into account the user’s repeated involvement with given topics, we can predict the distribution of trend durations as well as the thresholds in popularity that lead to their emergence within social media. Detailed measurements of datasets from Twitter confirm the validity of the model and its predictions.

💡 Research Summary

The paper addresses a fundamental question in the study of online social platforms: why do some topics not only become popular but also remain in the public agenda for extended periods? To answer this, the authors develop a stochastic dynamical model that explicitly incorporates two distinct user behaviors: (i) the first time a user posts about a topic (First‑Time Post, FTP) and (ii) subsequent posts about the same topic (Repeated Post, RP).

The model begins with the growth of FTPs. Let Nₜ denote the cumulative number of FTPs for a given topic at discrete time t. The authors assume a multiplicative update Nₜ = (1 + χₜ) Nₜ₋₁, where χₜ are small, positive, independent, identically distributed random variables with mean μ and variance 2σ². For small χₜ the product can be approximated by an exponential, yielding log Nₜ ≈ ∑₁ᵗχₛ + log N₀. By the Central Limit Theorem the sum of χ’s is normally distributed, so Nₜ follows a log‑normal distribution. This theoretical result matches empirical observations of Twitter data, where the distribution of cumulative FTP counts is indeed log‑normal.

Trend termination is defined through a “vitality” metric φₜ = Nₜ/Nₜ₋₁ = 1 + χₜ. If φₜ falls below a threshold θ₁ the topic ceases to be a trend. The probability of this event at any interval is p = Pr(χₜ < log θ₁) = F(log θ₁), where F is the cumulative distribution function of χ. Because χ’s are i.i.d., the duration L of a trend follows a geometric distribution with mean E(L) = 1/p − 1 = 1/F(log θ₁) − 1.

To capture the empirically observed phenomenon that users often post repeatedly about the same topic, the model adds RP. At each interval, the instantaneous numbers of FTPs and RPs are denoted FTPₜ and RPₜ, respectively. The ratio μₜ = (FTPₜ + RPₜ)/FTPₜ measures the “resonance” between users and the topic. The authors assume μₜ is uniformly distributed on