Predicting Continued Visibility of Music Category Videos in South Korean YouTube Trending Lists

Abstract

Public video counters are easy to obtain, but a large cumulative audience does not necessarily imply repeated representation in a regional trending archive. This study asks whether views, interaction rates and completed channel history predict continued visibility of Music category videos in the South Korean YouTube archive. The source contains 1,825 Music rows; excluding 19 rows with corrupted identifiers leaves 801 usable videos from November 2017 to June 2018. Duplicate and coverage-aware reconstruction yields 691 eligible first-appearance windows. The primary outcome is representation on at least three of seven calendar dates, including the first observed appearance. A secondary landmark task predicts next-day archive presence. Chronological development selects a channel-history random forest and a daily momentum random forest without sharing videos between fitting and later evaluation cohorts. On 213 later first appearances, the selected first forest achieves Brier 0.2000 and ROC AUC 0.687, versus 0.2239 and 0.500 for training prevalence. Its Brier improvement over prevalence has a positive paired channel-cluster interval. On 493 daily landmarks from 228 new videos, the forest achieves equal-video Brier 0.2264 and AUC 0.640; its Brier improvement over prevalence is 0.0145 with a 95 percent interval of 0.0057 to 0.0236. A pooled 20 percent monitoring allocation identifies 27 positive videos among 43 selected, versus 16 with view-count ranking; this historical comparison does not establish a causal promotion benefit. Comment and like rates have different conditional coefficient directions, and strict coverage restriction weakens first-appearance gains. The contribution is an auditable two-time-point forecasting and monitoring design, rather than a causal promotional claim or a model of the present YouTube Charts system.