Research · NewsShorts-Bench · Stage 13: Engagement prediction
Predicting YouTube 'most replayed' moments on news: position beats content
On 117 news videos, a position prior beats every content proxy (Spearman 0.143, HIT@15 0.222), though intervals overlap. Every proxy stays under 0.18 Spearman, so the pipeline uses no engagement model.
- Data
- 117 news videos
- Training
- 2,500 videos
- Compute
- CPU VM
- Metric
- Spearman
Rank correlation with replay scores on news videos (Spearman) · higher is better · dot = measured · bar = 95% CI
YouTube's "most replayed" graph is the only large-scale record of which seconds people come back to. If a cheap signal from the video itself (loud audio, a busy frame, a striking sentence) predicted that graph, a clipping pipeline could aim for those moments. On general YouTube, audio does best. On news, the best predictor is simply where in the video the moment is. And every cheap proxy is weak.
Read the full study on Every Local AI