Recommendation systems improve results by verifying retrieved items
Recommendation Retrievers Need Verifiers: Universal Generative Reranking for Sequential Recommendations
Information RetrievalArtificial Intelligence
Summary
Many modern recommendation systems pick a short list of items to show users, but sometimes good items are missed because they appear further down the list. The authors propose an additional step that rechecks the candidate items using a small, extra model, called a verifier, which decides if some of the skipped items should be promoted into the final list. This verifier is simple to add after the main recommendation model is trained and helps catch relevant items earlier without retraining the original model. Tests on Amazon and music recommendation data show improvements across various well-known recommendation methods.
What this means in practice
- •For ecommerce engineers: Boost final product recommendations by efficiently promoting relevant but initially lower-ranked candidates without retraining existing models.
- •For music streaming developers: Enhance music playlist suggestions by reranking candidate tracks with a lightweight verifier that improves recall on top recommendations.
Authors
Benyu Zhang, Qiang Zhang, Rui Li, Qunshu Zhang, Devansh Tandon, Neeraj Bhatia
Abstract
First-stage recommenders in multi-stage systems produce a ranked candidate list from which a limited prefix is forwarded to downstream rankers. Because each forwarded item must be processed by more expensive ranking stages, this shortlist cannot be arbitrarily large. The first-stage objective is therefore high coverage of relevant items within the forwarded prefix, commonly measured by Recall@$k$. A relevant item may be available deeper in the retrieved list but absent from the shorter prefix that is actually consumed. This paper studies post-hoc verification for promoting such candidates into the consumed shortlist without retraining or replacing the retriever. We introduce a lightweight generative verifier for retrieval models. Given a retriever state and a candidate item, the verifier scores the item through the likelihood of its identifier tokens. It is trained post hoc with next-token cross entropy, requires no sampled negatives or candidate pool during training, and scores only the retriever's top-$K$ candidates at inference. The interface is minimal: the retriever supplies a query state and candidate items, and the item representation can use any fixed tokenization. Across Amazon product recommendation and YaMBDa music recommendation, the same verifier training recipe improves Recall@10 for SASRec, GRU4Rec, NextItNet, and MiniOneRec. Ablations show that the improvements are not explained solely by injecting item-content features into the retriever, supporting verification as a post-hoc output-side adaptation mechanism.