Research map / Label-efficient and robust learning / Recognition and long-tail
Denoising Neural Reranker for Recommender Systems
Wenyu Mao, Shuchang Liu, HailanYang, Xiaobei Wang, Xiaoyu Yang, Xu Gao, Xiang Li, Lantao Hu, Han Li, Kun Gai, An Zhang, Xiang Wang
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 |
|---|---|---|---|
| replace direct use of retriever scores with denoising framework DNR (ablation) | PRM reranker · Kuaivideo | NDCG@6: 66.77-67.58 (score variants) -> 70.15 | Improved |
| feed retriever scores to reranker as extra input (concat, add, or weights) (ablation) | PRM reranker · Kuaivideo | NDCG@6: 55.93 -> 67.58 (c score) | Improved |
| apply DNR on a different reranker backbone (Pier) | Pier · Kuaivideo | NDCG@6: 65.11 -> 69.35 | Improved |
| add heuristic Gaussian noise augmentation (L_z) to PRM (ablation) | PRM · Kuaivideo | NDCG@6: 55.93 -> 67.34 | Improved |
| add adversarial noise objective L_adv (ablation) | DNR-G · Kuaivideo | NDCG@6: 68.70 -> 69.60 | Improved |
| add noise distribution regularization L_x (ablation) | DNR-G · Kuaivideo | NDCG@6: 69.10 -> 69.60 | Improved |
| add adversarial noise objective L_adv (ablation) | DNR-B · Kuaivideo | NDCG@6: 69.84 -> 70.15 | Improved |
| add noise distribution regularization L_x (ablation) | DNR-B · ML-1M | NDCG@6: 76.62 -> 77.12 | Improved |
| switch from heuristic Gaussian generator to model-based adversarial generator (ablation) | DNR · Amazon Book | NDCG@6: 81.32 -> 82.57 | Improved |
| switch from heuristic Beta generator to model-based adversarial generator (ablation) | DNR · Kuaivideo | NDCG@6: 68.71 -> 70.15 | Improved |
| use Beta noise instead of Gaussian noise (ablation) | DNR · Kuaivideo | NDCG@6: 69.60 -> 70.15 | Improved |
| increase lambda_c, lambda_m or lambda_e beyond optimum (ablation) | DNR · Kuaivideo | rerank performance | Hurt |
Closest papers by meaning
- UniRank: Unified List-wise Reranking via Confidence-Ordered Denoising (NeurIPS 2026 · similarity 0.89)
- Language Ranker: A Lightweight Ranking framework for LLM Decoding (NeurIPS 2025 · similarity 0.86)
- Advancing Loss Functions in Recommender Systems: A Comparative Study with a Rényi Divergence-Based Solution (AAAI 2025 · similarity 0.85)
- Contrastive Retrieval Heads for Improved Attention-Based Reranking (NeurIPS 2026 · similarity 0.85)
- On Efficiency-Effectiveness Trade-off of Diffusion-based Recommenders (NeurIPS 2025 · similarity 0.85)
- Ada-Retrieval: An Adaptive Multi-Round Retrieval Paradigm for Sequential Recommendations (AAAI 2024 · similarity 0.85)
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.
