Modelling of Trends in Twitter Using Retweet Graph Dynamics

Marijn ten Thij, Tanneke Ouboter, Daniël Worm, Nelly Litvak, Hans van den Berg, Sandjai BhulaiLecture Notes in Computer Science 2014 · DOI · arXiv · OpenAlex

Abstract

This paper proposes a novel random graph model to understand how Twitter trends emerge by tracking the structural evolution of retweet networks. We discover that a "densification" of the network's largest connected component—where the ratio of retweets to users exceeds 1.0—often serves as a predictive early-warning signal for a topic's peak popularity.

Introduction

Trend detection on social media has traditionally relied on content analysis, such as monitoring keyword frequency or sentiment shifts. While effective, these methods often overlook the topological dynamics that drive viral growth. This study proposes a structural approach, focusing on the formation and densification of retweet graphs. We aim to identify the mechanical signatures of trending topics by tracking how users interact and form a cohesive "giant component" over time.

Methodology

We model the growth of a retweet graph $G = (V, E)$ using a modified random graph framework. The model accounts for three primary events:

  1. T1 (New Component): A user posts an original message, creating a new tree.
  2. T2 (Tree Expansion): A new user retweets an existing message.
  3. T3 (Cycle/Merge): An existing user retweets another existing user, creating cycles or merging previously separate trees.

The model is parameterized by $\lambda$ (arrival rate), $p$ (new vs. existing user retweeting), and $q$ (the "superstar" preference for the original author). This allows us to theoretically derive the expected edge density and LCC size, providing a basis for parameter estimation from empirical data.

Empirical Analysis

The model was tested against 22 datasets spanning various event types, including the "Project X Haren" riot, major sports finals, and global tragedies. These datasets allow for a comparison of structural "fingerprints" across different social contexts.

Results: Densification as a Predictive Signal

The most significant observation is the relationship between network topology and tweet volume. In many cases, the retweet graph undergoes a phase transition before the trend reaches its maximum visibility.

Tweet volume for the Project X dataset, showing a clear peak on September 21st.

As shown in Figure 1, the "Project X" event saw a massive spike in activity on September 21st. However, structural analysis reveals that the network began to change much earlier.

Evolution of edge density and LCC size for Project X. The LCC densifies (red line in 'a' crosses 1.0) before the volume peak.

Figure 2 illustrates the evolution of the Largest Connected Component (LCC). Notably, the edge density of the LCC (red line in Fig 2a) crosses the critical threshold of 1.0 nearly 24 hours before the volume peak. This "densification" indicates that the conversation has moved from fragmented groups to a singular, highly interconnected community, serving as a precursor to viral spread.

Conclusion

By focusing on the dynamics of retweet graphs, we provide a theoretical and empirical framework for understanding trend emergence. The discovery of densification as an early-warning signal offers a powerful tool for predictive social media monitoring. Future research will refine the model by incorporating temporal decay and dynamic parameters to better capture long-term conversation lifecycles.

Summary updated

BibTeX
@inproceedings{thij2014modelling,
  author = {Marijn ten Thij and Tanneke Ouboter and Daniël Worm and Nelly Litvak and Hans van den Berg and Sandjai Bhulai},
  title = {{Modelling of Trends in Twitter Using Retweet Graph Dynamics}},
  booktitle = {Lecture Notes in Computer Science},
  year = {2014},
  doi = {10.1007/978-3-319-13123-8\_11},
}
Figure