Attention Surges and Recovery in Wikipedia Visits to Korean Music Artists

Abstract

Daily attention peaks are easy to notice, but a falling peak does not establish that attention has returned to its previous level. This study develops a transparent episode description that separates an observable surge alert, confirmed baseline return, residual attention burden and renewed attention. The data comprise 10,404 daily user-classified visits to six English Wikipedia artist pages from 1 January 2022 to 30 September 2026. A past-only detector combines a robust log-count deviation of at least three with a count at least twice the preceding 28-day median. Accepted onsets are separated by 29 days. Recovery requires three consecutive post-onset days at or below 1.25 times the frozen pre-alert median. Follow-up is limited to 28 days, with explicit right censoring. The detector identifies 92 episodes, of which 91 have complete follow-up. The pooled Kaplan–Meier median confirmed return is day 8. Among complete episodes, 42 return by day 7, 64 by day 14 and 78 by day 28. Updating the reference median every day raises the day-28 count to 90, showing that recovery classifications depend materially on reference adaptation. The median positive excess burden is 9.80 baseline equivalent days, and the median share accruing on days 8–28 is 37.5%. Ninety episodes confirm a half-excess decline after the first-week peak, but 12 of those still lack full baseline return by day 28. Renewed attention occurs in 14 of 73 episodes with a full seven-day observation opportunity after return. Rule perturbations and calendar-block resampling quantify descriptive sensitivity. The results support monitoring that records intensity, return status and residual burden separately. They describe page-visit dynamics within a selected historical panel and do not identify unique fans, event causes or promotional effects.