September 22, 2026
Minutes, hours or weeks: the time window decides which social media trends you see
A study of 14 verified campaigns shows that the length of the analysis window changes what is detected. It holds for coordination and for any trend analysis.
"Trend" does not mean the same thing if you look at the last hour or the last quarter. It sounds obvious, but few reports say which period they used and why. A paper by Georgios Panayiotou, Lorenzo Mannocci and Maurizio Tesconi, published on 18 September and accepted at the AIDEM 2026 workshop, puts numbers on that decision.
What they study
How coordinated behaviour is detected on social media, that is, groups of accounts sharing the same thing in a synchronised way. To detect it you have to set a time window. Do two accounts count as coordinated if they share the same thing ten seconds apart? And a day apart? The authors test different windows on 14 datasets of verified information operations.
What they find
- Short windows, from ten seconds to one minute: they capture highly synchronised coordination, with high precision.
- Long windows, from hours to days: they bring out slower, more diffuse patterns, but also more noise.
- The number of communities detected grows as the window gets longer.
- Overlap between windows barely matters: using consecutive, non-overlapping windows gives very similar results and costs less to compute.
The authors' conclusion is that window length is not a technical setting. It is a decision that determines which phenomenon you are observing.
Why it matters beyond bots
Because the same happens with any social media trend analysis. A sound that explodes on TikTok for 48 hours, a format that grows over three months and a shift in values that takes years are different things, and each is only visible at its own scale. Looking only at the week turns every peak into a trend. Looking only at the year erases them all.
How we read it at Trendia
Two simple rules:
- Look at more than one scale. What appears in hours is momentum. What holds for weeks starts to be a trend.
- Always say which window was used. The period should be written in the report, in plain sight, like the source of the data.
Before trusting a trend report, ask which period it covers. If it does not say, half the answer is missing.
You can see how we approach it in a real project.
Source: Setting the clock: Evaluating temporal window parameters for coordinated behavior detection, arXiv, 18 September 2026.