Recommender systems improve training with ranking-scored trial edits
Try Before You Evolve: Ranking-Scored Trials for Recursive Self-Improvement of Recommender Systems
Information Retrieval
Summary
Recommender systems need to adjust how they learn for different sets of data. The authors found that trying one change at a time and fully retraining the model is slow and often wastes effort. They created a method called RecRSI, which tests many small changes briefly before choosing the most promising one to fully train. This makes the system improve faster and better than older methods, especially for popular shopping datasets. It helps recommenders learn more efficiently by judging changes early on using a ranking score.
What this means in practice
- •For e-commerce platform engineers: Speed up and improve tuning of product recommendation models by testing multiple recipe changes early and selecting the best based on initial ranking scores.
- •For music streaming developers: Enhance music recommender efficiency by quickly evaluating potential training adjustments and committing only the most promising edits.
Authors
Xu Huang, Jiawei Feng, Jiancan Wu
Abstract
Recommenders on Web platforms such as e-commerce and music streaming need their training recipes re-tuned for each dataset. Large language model (LLM) agents can now rewrite these recipes, enabling recursive self-improvement (RSI) of recommender systems. Existing agents propose one edit per round, fully retrain the recommender, and keep the edit only if its ranking quality improves. Unlike programs or patches that can be scored by execution, a recommender edit changes how the model is trained, which raises two challenges. First, ranking feedback is delayed: an edit's ranking quality is known only after full retraining, so much of the retraining budget goes to edits that are later discarded. Second, the signals available before retraining are unreliable: sampled training losses can look solved while full-ranking performance remains poor, and they are not comparable across edits that change the loss or sampler. We propose RecRSI, which generates diverse candidate edits from a diagnosis of the current model and selects one for full retraining using early ranking feedback: each candidate receives a short trial, the first steps of its full training run, scored by full-ranking NDCG@10 on held-out probe users. With two sequential recommenders on three datasets and the same number of full retrainings, RecRSI significantly outperforms existing self-evolving agents in every setting, and on both Amazon datasets reaches in its first round a ranking quality that none of them reaches in five.