Mapping sampled Twitter activity to actual platform activity for time‑scale calibration

Ascertain a mapping from the activity volume observed in Twitter datasets obtained via the platform’s first API version (e.g., the Brexit and VaxNoVax datasets) to the corresponding actual platform activity at the time of collection, so as to relate the number of simulation iterations in the agent‑based model to real‑world time scales and enable principled parameter estimation.

Background

The study calibrates an agent‑based model to empirical cascade size distributions using datasets sampled from Twitter. Because these datasets represent only a small, time‑dependent sample of platform activity, the authors highlight a key limitation: the lack of a known mapping between sampled and actual activity volumes.

This prevents a straightforward alignment between simulation iterations and real‑world time, complicating parameter estimation for dynamics that evolve over time. Establishing such a mapping would improve model calibration and interpretability.

References

We cannot know how the volume of activity represented in the sample maps to the actual one in the system at the time. In essence, it is not trivial to match one simulation iteration to a time scale in the real world.

— Mechanistic interplay between information spreading and opinion polarization (2410.17151 - Oliveira et al., 22 Oct 2024) in Methods, Model calibration task (Section 3.5)