Research map / Reinforcement learning / Offline RL
A Model-Based Reinforcement Learning with Adversarial Training for Online Recommendation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use IRecGAN instead of PGU | Offline recommendation reranking · CIKM Cup 2016, random split | P@10 (%): 34.12$\pm$0.52 -> 35.06$\pm$0.48 | Helped |
| add adversarial training to the user behavior model (ablation) | LSTM · CIKM Cup 2016, random split | P@1 (%): 8.20$\pm$0.65 -> 8.55$\pm$0.63 | Helped |
| use IRecGAN instead of PG | Offline recommendation reranking · CIKM Cup 2016, split by session ID | P@10 (%): 32.44$\pm$1.16 -> \textbf{33.45}$\pm$0.71 | Helped |
| add adversarial training to the user behavior model (ablation) | LSTM · CIKM Cup 2016, split by session ID | P@1 (%): 9.68$\pm$0.29 -> \textbf{11.06}$\pm$0.23 | Helped |
| add adversarial training to the user behavior model (ablation) | LSTM · CIKM Cup 2016, split by time | P@1 (%): 7.67$\pm$0.12 -> \textbf{7.90}$\pm$0.14 | Helped |
| use IRecGAN instead of PGU | Offline recommendation reranking · CIKM Cup 2016, split by time | P@10 (%): 29.45$\pm$0.37 -> \textbf{30.07}$\pm$0.15 | Helped |
| train PG with importance sampling | PG · CIKM Cup 2016, random split | P@10 (%): 33.28$\pm$0.71 -> 28.13$\pm$0.45 | Hurt |
| learn from generated data without adversarial training (ablation) | PG · CIKM Cup 2016, random split | P@10 (%): 33.28$\pm$0.71 -> 34.12$\pm$0.52 | Helped |
| add adversarial training and discriminator-based reward scaling to PGU (ablation) | PGU · CIKM Cup 2016, random split | P@10 (%): 34.12$\pm$0.52 -> 35.06$\pm$0.48 | Helped |
| use IRecGAN instead of PG-online&offline | Online/offline policy learning · Simulated online learning | coverage@1 and coverage@10 | Helped |
| reuse generated data for offline policy-gradient learning (ablation) | PG-online · Simulated online learning | coverage@1 and coverage@10 | Helped |
| apply adversarial training to generate samples (ablation) | LSTM and IRecGAN · Simulated online test, when reward generation is estimable | coverage@r | Helped |
Closest papers by meaning
- Generative Adversarial User Model for Reinforcement Learning Based Recommendation System (ICML 2019 · similarity 0.92)
- Constrained Latent Action Policies for Model-Based Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.86)
- Adversarial Policy Optimization for Offline Preference-based Reinforcement Learning (ICLR 2025 · similarity 0.86)
- On the Unexpected Effectiveness of Reinforcement Learning for Sequential Recommendation (ICML 2024 · similarity 0.86)
- Distributional Off-Policy Evaluation for Slate Recommendations (AAAI 2024 · similarity 0.86)
- Off-Policy Actor-Critic for Adversarial Observation Robustness: Virtual Alternative Training via Symmetric Policy Evaluation (ICML 2025 · similarity 0.86)
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.