Research map / Video / Video segmentation and tracking
Relative Advantage Debiasing for Watch-Time Prediction in Short-Video Recommendation
Emily Liu, Kuan Han, Minfeng Zhan, Bocheng Zhao, Guanyu Mu, Yang Song
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use two-stage RAD-U instead of one-stage CQE (ablation) | RAD · KuaiRand-Pure (10 user clusters) | XGAUC: 0.7044 -> 0.7131 | Improved |
| use learned multiquantile embedding instead of CQE (ablation) | RAD-U · KuaiRand-Pure (10 user clusters) | Wasserstein distance: 9.3945 -> 2.2793 | Improved |
| replace exact empirical CDF with learned embedding (ablation) | RAD · KuaiRand-Pure | XGAUC: 0.7131 -> 0.7101 | No clear effect |
| fuse user and video sides (RAD-UV) vs RAD-U (ablation) | RAD · KuaiRand-Pure | XAUC: 0.7151 -> 0.7178 | Improved |
| fuse user and video sides (RAD-UV) vs RAD-V (ablation) | RAD · KuaiRand-Pure | MAE: 18.050 -> 18.050 | No clear effect |
| fuse user and video sides (RAD-UV) vs RAD-V (ablation) | RAD · KuaiRand-Pure | Video Group XAUC: 0.6803 -> 0.6786 | Hurt |
| RAD instead of D2Q | RAD · KuaiRand-Pure | MAE: 19.763 -> 18.050 | Improved |
| RAD instead of CQE | RAD · Offline industrial | Video Group XAUC: 0.594 -> 0.631 | Improved |
| deploy RAD-UV vs production model | RAD · Online A/B | Skip Rate: -0.8640% | Improved |
| deploy RAD-V vs production model | RAD · Online A/B | Watch Time: 0.0562% | No clear effect |
Closest papers by meaning
- CREAD: A Classification-Restoration Framework with Error Adaptive Discretization for Watch Time Prediction in Video Recommender Systems (AAAI 2024 · similarity 0.88)
- Calibrating Video Watch-time Predictions with Credible Prototype Alignment (ICML 2025 · similarity 0.87)
- Exploiting Fine-Grained Skip Behaviors for Micro-Video Recommendation (AAAI 2025 · similarity 0.84)
- Preference Diffusion for Recommendation (ICLR 2025 · similarity 0.83)
- Fading to Grow: Growing Preference Ratios via Preference Fading Discrete Diffusion for Recommendation (NeurIPS 2025 · similarity 0.82)
- Be Aware of the Neighborhood Effect: Modeling Selection Bias under Interference (ICLR 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
