Research map / Reinforcement learning / RL from feedback
Improving LLM-Based Recommenders with Conservative Generative Flow Networks
Xuan Yu, Feng Niu, Rui Zhu, Yudong Zhang, Xu Wang, Yang 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 Flower (SubTB) with conservative SubTB (CFlower) | LLM recommender · Amazon Video Games | NDCG@5: 0.0546 -> 0.0646 | Improved |
| replace Flower with CFlower | LLM recommender · Amazon Movies and TV | NDCG@5: 0.0956 -> 0.1036 | Improved |
| replace Flower with CFlower | LLM recommender · Amazon CDs and Vinyl | NDCG@5: 0.0702 -> 0.0737 | Improved |
| replace Flower with CFlower | LLM recommender · Amazon Video Games | HR@5: 0.0803 -> 0.0841 | Improved |
| replace Flower with CFlower | LLM recommender · Amazon CDs and Vinyl | exposure entropy H: 7.928 -> 8.132 | Improved |
| CFlower versus D3 (accuracy-oriented baseline) | CFlower · Amazon CDs and Vinyl | NDCG@5: 0.0810 vs 0.0737 | Hurt |
| replace Flower with CFlower for distribution matching | LLM recommender · target item distribution | Jaccard similarity: 0.799 -> 0.854 | Improved |
| replace Flower with CFlower for distribution matching | LLM recommender · target item distribution | total variation distance: 0.436 -> 0.332 | Improved |
| use CFlower instead of BIGRec as reference policy for S-DPO | S-DPO · Amazon CDs and Vinyl | NDCG@5: 0.0716 -> 0.0805 | Improved |
| use CFlower instead of Flower as reference policy for S-DPO | S-DPO · Amazon Video Games | NDCG@5: 0.0640 -> 0.0737 | Improved |
| use CFlower instead of Flower as reference policy for DMPO | DMPO · Amazon Movies and TV | NDCG@5: 0.0979 -> 0.1002 | Hurt |
| add offline constraints (dataset-constrained learning) (ablation) | SubTB/CSubTB · Amazon recommendation | NDCG/HR | Improved |
Closest papers by meaning
- Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations (ICML 2024 · similarity 0.88)
- Echoes in Filter Bubble: Diagnosing and Curing Popularity Bias in Generative Recommender Systems (NeurIPS 2026 · similarity 0.86)
- Revisiting Non-Acyclic GFlowNets in Discrete Environments (ICML 2025 · similarity 0.86)
- Beyond Squared Error: Exploring Loss Design for Enhanced Training of Generative Flow Networks (ICLR 2025 · similarity 0.85)
- Objective Shaping with Hard Negatives: Windowed Partial AUC Optimization for RL-based LLM Recommenders (NeurIPS 2026 · similarity 0.85)
- Can LLMs Outshine Conventional Recommenders? A Comparative Evaluation (NeurIPS 2025 · 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.
