STORM: Stepwise Token Optimization with Reward-Guided Beam Search

• URL: http://arxiv.org/abs/2606.10621v1
• Date: Tue, 09 Jun 2026 09:18:09 GMT
• Title: STORM: Stepwise Token Optimization with Reward-Guided Beam Search
• Abstract summary: STORM (Stepwise Token Optimization with Reward-guided beaM search) is a self-supervised framework for lexical query expansion.<n>It lets 0.6B-8B backbones match or surpass competitive LLM rewriters while retrieving as fast as plain BM25.<n>It further transfers zero-shot to 18 languages (MIRACL), beating dedicated multilingual dense retrievers on average.
• Score: 26.876242989882787
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Modern retrieval increasingly relies on dense and learned-sparse neural models that are effective but require encoding the entire corpus into a specialized index, rebuilt whenever the model changes. Lexical retrievers like BM25 stay efficient and transparent on a standard inverted index that need not change as models evolve, but suffer from vocabulary mismatch. LLM query rewriting can help, yet prompted rewriters emit well-formed but retrieval-ineffective or harmful-terms, and training against a retrieval reward gives only delayed, sequence-level supervision that obscures which terms helped. We introduce STORM (Stepwise Token Optimization with Reward-guided beaM search), a self-supervised framework for lexical query expansion. STORM trains the rewriter through generation guided by retrieval metrics: at each step, candidate expansions are scored against the BM25 index and low-reward continuations pruned, turning the retrieval reward into a token-level signal that concentrates exploration on retrieval-effective vocabulary. Across TREC DL and BEIR, STORM lets 0.6B-8B backbones match or surpass competitive LLM rewriters while retrieving as fast as plain BM25; at 8B it rivals far larger proprietary rewriters. It further transfers zero-shot to 18 languages (MIRACL), beating dedicated multilingual dense retrievers on average, making STORM a competitive, infrastructure-light alternative to dense neural retrieval.

Related papers

• SHIFT: Self-reconstruction Harnesses Implicit Fine-grained Thinking for Retrieval [31.80088730114538]
  We propose SHIFT, a retrieval training framework based on LLMs.<n>We transfer LLMs into reasoning-efficient retrievers with residual projection and task-oriented bidirectional attention aggregation in the latent space.<n>Experiments on reasoning-intensive retrieval benchmarks show that SHIFT consistently outperforms other widely used retrievers.
  arXiv  Detail & Related papers  (2026-07-23T14:00:09Z)
• Latent Terms: Dense Retrievers Contain Trivially Extractable BM25-ready Zipfian Vocabularies [1.938658977168075]
  We propose Latent Terms, a method revealing that models trained for dense retrieval learn representations that can trivially be decomposed into sparse features.<n>When trained on frozen retrievers, Sparse Autoencoders without any retrieval-specific adjustments extract a latent vocabulary with approximately Zipfian collection statistics.<n>Latent Terms is able to match or outperform single-vector scoring methods from its own base model as well as comparable SPLADE variants.
  arXiv  Detail & Related papers  (2026-05-28T05:36:37Z)
• Superintelligent Retrieval Agent: The Next Frontier of Information Retrieval [25.731213365755234]
  We introduce textitSuperIntelligent Retrieval Agent (SIRA)<n>SIRA can compress multi-round exploratory search into a single corpus-discriminative retrieval action.<n>It can exceed substantially more expensive multi-round search while remaining interpretable, training-free, and efficient.
  arXiv  Detail & Related papers  (2026-05-07T17:54:29Z)
• SAGE: Benchmarking and Improving Retrieval for Deep Research Agents [60.53966065867568]
  We introduce SAGE, a benchmark for scientific literature retrieval comprising 1,200 queries across four scientific domains, with a 200,000 paper retrieval corpus.<n>We evaluate six deep research agents and find that all systems struggle with reasoning-intensive retrieval.<n> BM25 significantly outperforms LLM-based retrievers by approximately 30%, as existing agents generate keyword-oriented sub-queries.
  arXiv  Detail & Related papers  (2026-02-05T18:25:24Z)
• LLMs as Sparse Retrievers:A Framework for First-Stage Product Search [103.70006474544364]
  Product search is a crucial component of modern e-commerce platforms, with billions of user queries every day.<n>Sparse retrieval methods suffer from severe vocabulary mismatch issues, leading to suboptimal performance in product search scenarios.<n>With their potential for semantic analysis, large language models (LLMs) offer a promising avenue for mitigating vocabulary mismatch issues.<n>We propose PROSPER, a framework for PROduct search leveraging LLMs as SParsE Retrievers.
  arXiv  Detail & Related papers  (2025-10-21T11:13:21Z)
• InsertRank: LLMs can reason over BM25 scores to Improve Listwise Reranking [3.1125398490785217]
  InsertRank is an LLM-based reranker that leverages lexical signals like BM25 scores during reranking to further improve retrieval performance.<n>With Deepseek-R1, InsertRank achieves a score of 37.5 on the BRIGHT benchmark, and 51.1 on the R2MED benchmark, surpassing previous methods.
  arXiv  Detail & Related papers  (2025-06-17T01:04:45Z)
• Mistral-SPLADE: LLMs for better Learned Sparse Retrieval [7.652738829153342]
  We propose to use decoder-only model for learning semantic keyword expansion. We use Mistral as the backbone to develop our Learned Sparse Retriever similar to SPLADE. Our experiments support the hypothesis that a sparse retrieval model based on decoder only large language model (LLM) surpasses the performance of existing LSR systems.
  arXiv  Detail & Related papers  (2024-08-20T18:21:54Z)
• Query Rewriting for Retrieval-Augmented Large Language Models [139.242907155883]
  Large Language Models (LLMs) play powerful, black-box readers in the retrieve-then-read pipeline. This work introduces a new framework, Rewrite-Retrieve-Read instead of the previous retrieve-then-read for the retrieval-augmented LLMs.
  arXiv  Detail & Related papers  (2023-05-23T17:27:50Z)
• Large Language Models are Strong Zero-Shot Retriever [89.16756291653371]
  We propose a simple method that applies a large language model (LLM) to large-scale retrieval in zero-shot scenarios. Our method, the Language language model as Retriever (LameR), is built upon no other neural models but an LLM.
  arXiv  Detail & Related papers  (2023-04-27T14:45:55Z)
• Noise-Robust Dense Retrieval via Contrastive Alignment Post Training [89.29256833403167]
  Contrastive Alignment POst Training (CAPOT) is a highly efficient finetuning method that improves model robustness without requiring index regeneration. CAPOT enables robust retrieval by freezing the document encoder while the query encoder learns to align noisy queries with their unaltered root. We evaluate CAPOT noisy variants of MSMARCO, Natural Questions, and Trivia QA passage retrieval, finding CAPOT has a similar impact as data augmentation with none of its overhead.
  arXiv  Detail & Related papers  (2023-04-06T22:16:53Z)
• Bridging the Training-Inference Gap for Dense Phrase Retrieval [104.4836127502683]
  Building dense retrievers requires a series of standard procedures, including training and validating neural models. In this paper, we explore how the gap between training and inference in dense retrieval can be reduced. We propose an efficient way of validating dense retrievers using a small subset of the entire corpus.
  arXiv  Detail & Related papers  (2022-10-25T00:53:06Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.