Research map / Reinforcement learning / Visual and representation RL
Generative Adversarial User Model for Reinforcement Learning Based Recommendation System
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 GAN-LSTM instead of GAN-PW for the history embedding (ablation) | GAN user model · MovieLens, Last.fm, Yelp, Taobao, YooChoose, Ant Financial | Prec@1 (%): 66.6($\pm$0.7) -> 67.4($\pm$0.5); 24.1($\pm$0.8) -> 24.0($\pm$0.9); 72.0($\pm$0.2) -> 73.0($\pm$0.2); 34.7($\pm$0.6) -> | Mixed |
| Use L2 regularization instead of Shannon entropy (ablation) | GAN-LSTM user model · Yelp, Split 1 | Prec@1 (%): 73.1 -> 73.5 | Helped |
| Use L2 regularization instead of Shannon entropy (ablation) | GAN-LSTM user model · Yelp, Split 2 | Prec@1 (%): 72.8 -> 78.8 | Helped |
| Use L2 regularization instead of Shannon entropy (ablation) | GAN-LSTM user model · Yelp, Split 3 | Prec@1 (%): 73.1 -> 76.1 | Helped |
| Use cascading Q-networks instead of the additive Q-network (ablation) | GAN-based RL recommendation · Ant Financial test users, k=3 | cumulative reward: 23.60($\pm$1.06) -> 24.05($\pm$0.98) | Helped |
| Use cascading Q-networks instead of the additive Q-network (ablation) | GAN-based RL recommendation · Ant Financial test users, k=5 | cumulative reward: 23.19($\pm$1.17) -> 25.36($\pm$1.10) | Helped |
| Train cascading Q-networks with learned reward rather than ±1 reward (ablation) | GAN-based RL recommendation · Ant Financial test users, k=3 | CTR: 0.68($\pm$0.03) -> 0.74($\pm$0.03) | Helped |
| Train cascading Q-networks with learned reward rather than ±1 reward (ablation) | GAN-based RL recommendation · Ant Financial test users, k=5 | CTR: 0.78($\pm$0.03) -> 0.77($\pm$0.03) | No clear effect |
| Use a learned continuous reward instead of ±1 click reward (ablation) | GAN-based RL recommendation · Ant Financial test users, k=5 | cumulative reward: 25.15($\pm$1.04) -> 25.36($\pm$1.10) | No clear effect |
| Use RL optimization instead of the greedy policy | GAN-based recommendation · Ant Financial test users, k=3 | cumulative reward: 21.37($\pm$1.24) -> 24.05($\pm$0.98) | Helped |
| Use RL optimization instead of the greedy policy | GAN-based recommendation · Ant Financial test users, k=5 | cumulative reward: 22.97($\pm$1.22) -> 25.36($\pm$1.10) | Helped |
| Pre-train CDQN using the GAN user model before adapting to a new user | CDQN recommendation policy · New Ant Financial users | CTR | Helped |
Closest papers by meaning
- LLM-Powered User Simulator for Recommender System (AAAI 2025 · similarity 0.88)
- On the Unexpected Effectiveness of Reinforcement Learning for Sequential Recommendation (ICML 2024 · similarity 0.86)
- The Adaptive Q-Network for Recommendation Tasks with Dynamic Item Space (AAAI 2025 · similarity 0.86)
- Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations (ICML 2024 · similarity 0.85)
- Would You Like Your Data to Be Trained? A User Controllable Recommendation Framework (AAAI 2024 · similarity 0.84)
- MTRec: Learning to Align with User Preferences via Mental Reward Models (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.