Research map / Reinforcement learning / Policy gradient and actor-critic
RaCT: Toward Amortized Ranking-Critical Training For Collaborative Filtering
Sam Lobel*, Chunyuan Li*, Jianfeng Gao, Lawrence Carin
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 |
|---|---|---|---|
| Apply RaCT to VAE | VAE · ML-20M | NDCG@100: 0.4258 -> 0.4339 | Improved |
| Apply RaCT to VAE | VAE · Netflix | NDCG@100: 0.385 -> 0.392 | Improved |
| Apply RaCT to VAE | VAE · MSD | NDCG@100: 0.310 -> 0.319 | Improved |
| Replace the multinomial likelihood with a Gaussian likelihood (ablation) | VAE actor + RaCT · ML-20M | NDCG@100: 0.4339 -> 0.4224 | Hurt |
| Remove KL regularization and use a deterministic actor (ablation) | VAE actor + RaCT · ML-20M | NDCG@100: 0.4339 -> 0.4255 | Hurt |
| Use a linear encoder and decoder (ablation) | VAE actor + RaCT · ML-20M | NDCG@100: 0.4339 -> 0.4162 | Hurt |
| Remove the held-out interaction count feature |H0| (ablation) | Feature-based critic · ML-20M | NDCG@100 | Hurt |
| Remove the observed interaction count feature |H1| (ablation) | Feature-based critic · ML-20M | NDCG@100 | Hurt |
| Remove both interaction-count features (ablation) | Feature-based critic · ML-20M | NDCG@100 | Hurt |
| Train the critic with Recall@100 instead of NDCG@100 (ablation) | RaCT · ML-20M | Recall@20: 0.40269 -> 0.40316 | Improved |
| Train the critic with Recall@100 instead of NDCG@100 (ablation) | RaCT · ML-20M | NDCG@100: 0.43395 -> 0.43392 | No clear effect |
| Apply RaCT to WARP | WARP actor · ML-20M | NDCG@100: 0.3123 -> 0.3439 | Improved |
Closest papers by meaning
- Objective Shaping with Hard Negatives: Windowed Partial AUC Optimization for RL-based LLM Recommenders (NeurIPS 2026 · similarity 0.87)
- ConvRec-R1: Training LLM-based Conversational Recommender Systems with Reinforcement Learning (ICLR 2026 · similarity 0.86)
- From Pairwise to Ranking: Climbing the Ladder to Ideal Collaborative Filtering with Pseudo-Ranking (AAAI 2025 · similarity 0.85)
- On the Unexpected Effectiveness of Reinforcement Learning for Sequential Recommendation (ICML 2024 · similarity 0.85)
- Adaptive Hardness Negative Sampling for Collaborative Filtering (AAAI 2024 · similarity 0.85)
- Think before Recommendation: Autonomous Reasoning-enhanced Recommender (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.
