Research map / Learning theory and algorithms / Prediction and decision losses
Cache-Augmented Inbatch Importance Resampling for Training Recommender Retriever
Jin Chen, Defu Lian, Yucheng Li, Baoyun Wang, Kai Zheng, Enhong Chen
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 |
|---|---|---|---|
| add item-popularity proposal correction to inbatch sampling (SSL-Pop vs SSL) | SSL · Gowalla | NDCG@10: 0.1381 -> 0.1479 | Improved |
| add item-popularity proposal correction (SSL-Pop vs SSL) | SSL · Ta-feng | NDCG@10: 0.0437 -> 0.0625 | Improved |
| add uniformly sampled out-of-batch negatives (MNS vs SSL-Pop) | SSL-Pop · Echonest | NDCG@10: 0.1609 -> 0.1648 | Improved |
| replace inbatch popularity correction with importance resampling (BIR vs SSL-Pop) | SSL-Pop · Gowalla | NDCG@10: 0.1479 -> 0.1523 | Improved |
| replace inbatch popularity correction with importance resampling (BIR vs best baseline) | best baseline · Ta-feng | NDCG@10: 0.0634 -> 0.0675 | Improved |
| add cache of frequently sampled items to BIR (XIR vs BIR) (ablation) | BIR · Amazon | NDCG@10: 0.0833 -> 0.0877 | Improved |
| add cache of frequently sampled items to BIR (XIR vs BIR) (ablation) | BIR · Echonest | NDCG@10: 0.1682 -> 0.1842 | Improved |
| add cache of frequently sampled items to BIR (XIR vs BIR) (ablation) | BIR · Gowalla | RECALL@10: 0.1157 -> 0.1169 | Improved |
| add cache of frequently sampled items to BIR (XIR vs BIR) (ablation) | BIR · Tmall | NDCG@10: 0.0590 -> 0.0658 | Improved |
| BIR vs SSL-Pop on Gowalla recall (G-Tower stronger) | BIR · Gowalla | RECALL@10: 0.1176 vs 0.1157 | No clear effect |
| add item features (SellerID, CateID) with a two-layer item tower, BIR vs SSL-Pop | SSL-Pop · Tmall | NDCG@10: 3.06% | Improved |
| add item features, XIR vs SSL-Pop | SSL-Pop · Tmall | NDCG@10: 4.87% | Improved |
Closest papers by meaning
- Influential Exemplar Replay for Incremental Learning in Recommender Systems (AAAI 2024 · similarity 0.85)
- PSL: Rethinking and Improving Softmax Loss from Pairwise Perspective for Recommendation (NeurIPS 2024 · similarity 0.85)
- On Softmax Direct Preference Optimization for Recommendation (NeurIPS 2024 · similarity 0.84)
- LIME: Link-based User-item Interaction Modeling with Decoupled XOR Attention for Efficient Test Time Scaling (NeurIPS 2026 · similarity 0.84)
- Advancing Loss Functions in Recommender Systems: A Comparative Study with a Rényi Divergence-Based Solution (AAAI 2025 · similarity 0.84)
- On Efficiency-Effectiveness Trade-off of Diffusion-based Recommenders (NeurIPS 2025 · similarity 0.84)
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.
