Research map / Large language models / Language, safety and interpretability
Think Wise, Collaborate Effectively: A Rationale-Aware LLM-Based Recommender with Reinforcement Learning from Collaborative Signals
Chung Park, Taesan Kim, Hyeongjun Yun, Dongjoon Hong, Junui Hong, Kijung Park, MinCheol Cho, Min Sung Choi, Jihwan Seok, Jaegul Choo
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 collaborative-guided reward (CF confidence) on top of item-matching and formatting rewards in GRPO (ablation) | TWiCE-Rec · Amazon Musical Instruments and Health (HR@1, NDCG@5) | HR@1, NDCG@5 | Improved |
| add Stage 2 instruction tuning before GRPO (ablation) | TWiCE-Rec · Amazon Musical Instruments and Health | HR@1, NDCG@5 | Improved |
| add GRPO (Stage 3) after instruction tuning (ablation) | TWiCE-Rec · Amazon Musical Instruments and Health | HR@1, NDCG@5 | Improved |
| apply RL only to the base model without Stage 2 (ablation) | Gemma-3-1B-it · Amazon Musical Instruments and Health | HR@1, NDCG@5 | Mixed |
| increase CF confidence weight lambda up to 3 (ablation) | TWiCE-Rec collaborative-guided reward · Amazon Fashion and Clothing | HR@1, HR@5, NDCG@5 | Improved |
| increase lambda beyond 3 (ablation) | TWiCE-Rec collaborative-guided reward · Amazon Fashion and Clothing | HR@1, HR@5, NDCG@5 | Hurt |
| generate rationale at inference (ablation) | TWiCE-Rec · eight Amazon domains | HR@1 | Improved |
| train with rationale supervision (ablation) | TWiCE-Rec · eight Amazon domains (item-only inference) | HR@1 | Improved |
| use TWiCE-Rec versus best baseline on HR@1 | TWiCE-Rec · Amazon Fashion | HR@1: +63.31% | Improved |
| use TWiCE-Rec versus best baseline on HR@5 | TWiCE-Rec · Amazon Industrial and Scientific | HR@5: 23.94 -> 23.16 | Hurt |
| use TWiCE-Rec in online serving versus ML model | TWiCE-Rec · online A/B test | CTR: 10.67% -> 11.49% | Improved |
| use larger LLM backbone (1B, 4B, 12B) (ablation) | TWiCE-Rec · Amazon domains | recommendation performance | Improved |
Closest papers by meaning
- ConvRec-R1: Training LLM-based Conversational Recommender Systems with Reinforcement Learning (ICLR 2026 · similarity 0.91)
- Re2LLM: Reflective Reinforcement Large Language Model for Session-based Recommendation (AAAI 2025 · similarity 0.91)
- CoRA: Collaborative Information Perception by Large Language Model’s Weights for Recommendation (AAAI 2025 · similarity 0.90)
- Reinforced Latent Reasoning for LLM-based Recommendation (ICLR 2026 · similarity 0.90)
- R$^2$ec: Towards Large Recommender Models with Reasoning (NeurIPS 2025 · similarity 0.90)
- Bridging the User-side Knowledge Gap in Knowledge-aware Recommendations with Large Language Models (AAAI 2025 · similarity 0.90)
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.
