# Diagnosing and Mitigating Retrieval Bottlenecks in LLM-Based Cold-Start Recommendation

arXiv: https://arxiv.org/abs/2606.29947 · HTML: https://mlsanity.com/paper/2606.29947/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 34 references that are in the map.
- First posted on arXiv 29 Jun 2026.

## Builds on (most cited first, 22 of 22)

- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Dense Passage Retrieval for Open-Domain Question Answering](https://mlsanity.com/paper/2004.04906/) (2020)
- [Text Embeddings by Weakly-Supervised Contrastive Pre-training](https://mlsanity.com/paper/2212.03533/) (2022)
- [Self-Attentive Sequential Recommendation](https://mlsanity.com/paper/1808.09781/) (2018)
- [LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation](https://mlsanity.com/paper/2002.02126/) (2020)
- [M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation](https://mlsanity.com/paper/2402.03216/) (2024)
- [Neural Graph Collaborative Filtering](https://mlsanity.com/paper/1905.08108/) (2019)
- [C-Pack: Packed Resources For General Chinese Embeddings](https://mlsanity.com/paper/2309.07597/) (2023)
- [Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5)](https://mlsanity.com/paper/2203.13366/) (2022)
- [A Survey on Large Language Models for Recommendation](https://mlsanity.com/paper/2305.19860/) (2023)
- [TALLRec: An Effective and Efficient Tuning Framework to Align Large Language Model with Recommendation](https://mlsanity.com/paper/2305.00447/) (2023)
- [Large Language Models are Zero-Shot Rankers for Recommender Systems](https://mlsanity.com/paper/2305.08845/) (2023)
- [Is ChatGPT a Good Recommender? A Preliminary Study](https://mlsanity.com/paper/2304.10149/) (2023)
- [Are We Really Making Much Progress? A Worrying Analysis of Recent Neural Recommendation Approaches](https://mlsanity.com/paper/1907.06902/) (2019)
- [VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback](https://mlsanity.com/paper/1510.01784/) (2015)
- [How Can Recommender Systems Benefit from Large Language Models: A Survey](https://mlsanity.com/paper/2306.05817/) (2023)
- [MeLU: Meta-Learned User Preference Estimator for Cold-Start Recommendation](https://mlsanity.com/paper/1908.00413/) (2019)
- [Warm Up Cold-start Advertisements: Improving CTR Predictions via Learning to Learn ID Embeddings](https://mlsanity.com/paper/1904.11547/) (2019)
- [A Troubling Analysis of Reproducibility and Progress in Recommender Systems Research](https://mlsanity.com/paper/1911.07698/) (2019)
- [RecRanker: Instruction Tuning Large Language Model as Ranker for Top-k Recommendation](https://mlsanity.com/paper/2312.16018/) (2023)
- [Diagnosing LLM-based Rerankers in Cold-Start Recommender Systems: Coverage, Exposure and Practical Mitigations](https://mlsanity.com/paper/2604.16318/) (2026)

## Abstract (arXiv)

Large language models (LLMs) are increasingly used as rerankers in recommender systems, with the expectation that semantic understanding will help in cold-start and long-tail regimes. We test this assumption with a five-domain benchmark that explicitly separates reranking quality from retrieval coverage. In a positive-controlled regime where the gold item is guaranteed present, calibrated LLM rerankers fail to consistently outperform strong collaborative and content baselines under natural traffic, and within-family scaling from Qwen3-8B to Qwen3-32B narrows but does not close the gap on most domains. In a retrieval-realistic regime where the gold item is not injected, the bottleneck is more severe: standard single retrievers place the gold item in a 200-item pool only 4.6-22.9% of the time, largely because 32-91% of cold-start targets are brand-new items with no training interactions. We introduce LHF, a validation-trained learned hybrid fusion layer over a multi-retriever union pool, as a retrieval-side realizability baseline. LHF is the only combiner we test that beats every single retriever on all five domains and recovers 17-61% of oracle coverage headroom on content-rich domains, but only 5-7% on collaboratively strong domains. End-to-end experiments reveal the remaining mismatch: learned non-LLM ranking exploits the LHF pool, while prompt-level LLM reranking often degrades it. LLMs exhibit pockets of semantic cold-start advantage, especially in text-rich domains when the item is already present, but this advantage is largely unreachable in current retrieve-then-rerank pipelines. We release the benchmark protocol, splits, prompts, evaluation tooling, and archived reproducibility artifacts: data at https://doi.org/10.5281/zenodo.20991039 and code at https://doi.org/10.5281/zenodo.20993306.