Figure 4. Topical consistence over time despite user turnover.

We track the similarity D_RC of the generalist cores of the network in time (C_t) with respect to a fixed reference (the core of the network when the maximum of the nestedness is observed, C_max). Results for different w (12 h in the top panel; 3 days in the lower) show that only hashtags build a stable core, guaranteeing the semantic coherence of the topic across time; whereas the core of users suffers a high rate of turnover, indicating that users are frequently pushed to and from the periphery of the network.