Beyond Successor Accuracy: State Retention for Recursive Self-Improvement in Recommendation

2026-10-05

AI Summary

This study addresses the limitation in recursive self-improvement for recommender systems, where evaluating only the latest model overlooks complementary decisions across generations. To this end, it introduces the concept of "distributed progress," formulating state retention as an independent optimization problem. Methodologically, leveraging sequential encoders such as GRU4Rec, SASRec, and FMLP, the approach quantifies inter-model relationships via cross-generational advantages and predicts optimal retention strategies using a label-free rank separation statistic. Experimental results demonstrate that this method accurately identifies stronger model families under most settings, significantly outperforming direct successor models. These findings reveal that progress can reside within inter-generational relationships rather than relying solely on a single successor, effectively validating the superiority of cross-generational retention strategies.

Abstract

Recommendation recursive self-improvement (Rec-RSI) feeds recommender outputs into subsequent training. Evaluating each round solely through its latest model assumes that the successor consolidates the update, although pre- and post-update models may retain complementary ranking decisions. We term this \emph{distributed progress} and quantify it using cross-generation advantage (CGA), a marginally matched contrast between cross- and within-generation model pairs. A rank-separation statistic, label-free at selection time, predicts which family to retain. Across four datasets and three sequential recommendation encoders, the preferred retention regime varies by architecture: cross-generation pairing benefits GRU4Rec and SASRec, whereas FMLP initially favors within-generation pairing and shifts toward cross-generation pairing after a second update. Rank separation selects the stronger family in 12/12 first-update and 5/6 second-update dataset-encoder settings; on held-out tests, the selected family outperforms the direct successor in 34/36 trajectories. Five transfer mechanisms do not consistently reproduce these gains in one model. These findings establish state retention as a distinct Rec-RSI problem: progress may reside in relations between generations as well as in the latest model. Code is available at \href{https://github.com/Jinfeng-Xu/RecRSI}{https://github.com/Jinfeng-Xu/RecRSI}.

Problem

• Recommendation recursive self-improvement
• State retention
• Distributed progress
• Cross-generation advantage
• Rank separation

Innovation

• Recursive Self-Improvement
• Distributed Progress
• Cross-Generation Advantage
• Rank Separation
• State Retention