Generative recommenders improve long-tail item suggestions with memorization fix
Generative Recommenders Forget the Long Tail: Restoring Memorization in Retrieval and Ranking
Information Retrieval
Summary
Generative recommenders try to suggest items by generating item identifiers but tend to miss rare or less popular items, called the long tail, because of how they search and score options. The authors found this happens twice: during the selection and ranking steps. They created a simple add-on called CAPE that checks for items often bought or used together, helping the model remember and include those rare items better. This improves recommendation accuracy for niche products without slowing down the system much.
What this means in practice
- •For e-commerce platform engineers: Enhance product recommendation algorithms to include rare items customers like, increasing diversity and personalization without hurting overall accuracy.
- •For music streaming service developers: Improve recommendation systems to better surface niche or less popular tracks by combining generative models with co-occurrence information.
Authors
Xu Huang, Jiancan Wu, Hongzheng Qin, Jiawei Feng, Can Ye, Wencong Zeng, Fengze Li, Xiang Wang
Abstract
Generative recommenders replace the retrieve-then-rank cascade with a single model: beam search over semantic item identifiers performs retrieval, and the same likelihood performs ranking. We show that this shared score makes them forget long-tail items twice. In retrieval, beam search keeps identifier prefixes by the probability mass of their subtrees, so tail items are pruned early, and a wider beam admits few more of them. In ranking, the likelihood places tail items below head items even when they are admitted, so expanding retrieval alone brings little or no gain. Both failures reflect missing item-level memorization (i.e., direct recall of which items co-occur), which collaborative filtering provides and shared-prefix generation lacks. To this end, we propose CAPE (Co-occurrence-Aware Pool Expansion), a training-free plug-in that restores memorization at both stages: (i) a memorization channel retrieves co-occurrence-supported tail items outside the beam and scores them with the frozen generator, and (ii) memorization-calibrated ranking fuses the generator likelihood with co-occurrence evidence while keeping the generator's first item. Experiments with two generative backbones on product and music domains show that CAPE significantly improves long-tail accuracy over the backbones and over re-ranking baselines that share its candidates, preserves or improves overall accuracy, and adds a small fraction of the latency.