The 2020 bot label change causes half of the apparent 2016–2025 decrease in English Wikipedia attention time after 4,547 notable deaths, and the time did not decrease after mid-2020
Question: did attention to English Wikipedia articles after the deaths of notable people become shorter from 2016 to 2025? Data: a random sample per year of people in Wikidata who died in 2016–2025 and have 10 or more sitelinks (9,043 people; 4,547 after quality filters), with daily pageviews from 90 days before to 300 days after the death. Measure: T90, the days from the peak until 90% of the 90-day excess views above the pre-death baseline. Results: with the published 'user' label, median T90 decreased from 20 days (2016–2019) to 15 days (2021–2025), a trend of −2.7% per year (95% CI −3.6 to −1.9; errors clustered by month of death). On 2020-04-29 Wikimedia started to label some bot traffic 'automated'; before, it was 'user'. For all 2,292 sample deaths after June 2020 we added the automated views back. This made T90 longer by 8.4% (CI 7.6–9.4%). On this label-consistent series the median T90 decreased from 20 to 17 days (difference −3, CI −4 to −1) and the trend was −1.35% per year (CI −2.2 to −0.5), half of the user-label trend. The lowest values were in 2020–2021, and after June 2020 T90 did not decrease (+1.9% per year, CI −0.6 to +4.4). Larger peaks relative to the baseline explain most of the remaining trend: with peak size, baseline popularity, sitelinks and age as controls, the trend was −0.5% per year (CI −1.4 to +0.4). The next-day-to-peak ratio (0.46 in 2016, 0.44 in 2025) and the attention level 1 and 6–9 months later had no trend. Thus death bursts did not show the slower decay that an earlier SCI study found for top-list bursts. Limitations: the correction assumes that post-2020 'automated' traffic is the same kind of traffic that was in 'user' before 2020; English Wikipedia only; people with 10 or more sitelinks only; the follow-up analyses were added after the first results.
collective attention · Wikipedia pageviews · attention decay · death · Wikidata · bot traffic · data artefact · acceleration of attention · collective memory
References
7- doi:10.1038/s41467-019-09311-w
Reports accelerating collective attention on several platforms but not clearly on Wikipedia in 2012–2017; we test the question for death-triggered attention in 2016–2025.
- English Wikipedia top-list attention became slower, not faster, from 2016 to 2025: longer top-50 stays, lower turnover and slower burst decay in daily user pageview data sci:sha256:d88df5ed97e71c69813236e7fc57a74920fb497209326818f8096ed355bf8cdc
Found slower attention bursts in English Wikipedia top lists in 2016–2025; we compute three of its burst measures for death bursts and find no such slowing.
- doi:10.1038/s41562-018-0474-5
Models the decay of collective attention and memory, including Wikipedia biographies.
- doi:10.1038/s41598-022-25840-9
Two-phase decay model of Wikipedia attention after events, including deaths.
- doi:10.1073/pnas.2106152118
Measures attention to public figures after death in news and Twitter.
- doi:10.1126/sciadv.1602368
Uses Wikipedia pageviews to study collective memory.
- doi:10.1016/j.joi.2015.07.006
Reports faster decay of attention to scientific papers over time.
Cited by
5 publicationsPublic catalog entries on Sciscape.
- Against matched living controls, English Wikipedia views after 17,866 deaths in 2019–2023 were 8% higher in year 2 and not higher in year 4; year-4 change was +40% below age 60 and −11% at 85+
Shows the 2020 automated-traffic label change in Wikipedia user views; our matched design compares treated and control units in the same calendar months.
- usesThe 2016–2025 steepening of power-law attention decay after 5,434 notable deaths on English Wikipedia disappears when the 2020 'automated' traffic label change is undone
We use its label-consistency correction and its retained automated-view series for 1,395 of our 3,186 post-June-2020 deaths; our result supports its conclusion for a power-law exponent measure.
- Death gave a 2.4-fold year-1 rise in English Wikipedia views but no clear median change in years 2–4 against matched living people (1,517 deaths, 2016–2021); effects fell with age
Documents the 2020 change of the Wikimedia 'automated' traffic label, which affects user pageview baselines.
- The curated weekly English Wikipedia Top 25 Report shows no faster turnover in 2016–2026, and its 2013–2017 increase matches source changes and the loss of perennial pages
Related result on the 2020 change of the Wikimedia bot label, a data-source change of the same kind that we test in the Top 25 Report.
- The curated weekly English Wikipedia Top 25 Report shows no faster turnover in 2016–2026, and its 2013–2017 increase matches source changes and the loss of perennial pages
Related result on the 2020 change of the Wikimedia bot label, a data-source change of the same kind that we test in the Top 25 Report.