Reinforcement Learning with Conformal Action Sets: An Application to Sequential Recommendation
arXiv cs.AI · 7 Oct 2026, 03:40 AEST
Reinforcement Learning with Calibrated Pruning (RLCP) adapts the action set in sequential recommendation systems based on critic scores and an online threshold. Experiments show that RLCP variants achieve significantly higher catalog diversity compared to traditional methods.
Key facts
- Sequential recommenders typically use a fixed slate size even though the number of useful alternatives changes within a session.
- We propose Reinforcement Learning with Calibrated Pruning (RLCP), which adapts the retained action set using critic scores and an online threshold.
- Experiments on KuaiRand-Pure and MovieLens 1M compare two RLCP implementations with four RL baselines.
- In each of the 19 configurations, at least one RLCP variant achieves the highest catalog diversity, reaching $1.11\times$ to $5.21\times$ that of the strongest baseline.
Read aloud
Reinforcement Learning with Calibrated Pruning RLCP adapts the action set in sequential recommendation systems based on critic scores and an online threshold. Experiments show that RLCP variants achieve significantly higher catalog diversity compared to traditional methods. Sequential recommenders typically use a fixed slate size even though the number of useful alternatives changes within a session. We propose Reinforcement Learning with Calibrated Pruning RLCP, which adapts the retained action set using critic scores and an online threshold. Experiments on KuaiRand-Pure and MovieLens 1M compare two RLCP implementations with four RL baselines. In each of the nineteen configurations, at least one RLCP variant achieves the highest catalog diversity, reaching one dollars and eleven cents\times$ to five dollars and twenty one cents\times$ that of the strongest baseline.
Sources in this cluster
- Reinforcement Learning with Conformal Action Sets: An Application to Sequential Recommendation
arXiv cs.AI · 7 Oct 2026, 03:40 AEST