Search papers, labs, and topics across Lattice.
This paper introduces Reinforcement Learning with Calibrated Pruning (RLCP) for sequential recommendation, addressing the limitation of fixed slate sizes by adaptively pruning the action set based on critic scores and an online threshold derived from binary feedback. RLCP provides a theoretical guarantee on the proxy miss rate and decomposes value loss into filtering and selection components, enabling finite session reward bounds even with approximate critics and non-converged learning parameters. Experimental results demonstrate RLCP's ability to significantly increase catalog diversity (1.11x to 5.21x) compared to strong baselines, while maintaining competitive session depth and set sizes.
Sequential recommenders can dramatically boost catalog diversity by adaptively pruning action sets, achieving up to 5.21x diversity gains without sacrificing session depth.
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. The threshold is updated from binary feedback indicating whether the set contains an action in a proxy target. We prove a deterministic bound on the observed proxy miss rate along adaptive trajectories. To quantify the effect of pruning on reward, we derive an exact decomposition of value loss into filtering and selection losses. Under explicit proxy and critic approximation conditions, this decomposition yields a finite session reward bound that also accounts for imperfect selection and set truncation, without requiring the learning parameters to converge. 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, with competitive session depth and no larger retained sets.
