Query Rewriting — References

Query reformulation as a retrieval technique, from classical relevance feedback to generation-based reformulation.

Classical reformulation

• Relevance Feedback in Information Retrieval — J. J. Rocchio, in The SMART Retrieval System: Experiments in Automatic Document Processing, 1971. The original formulation of moving a query toward relevant documents and away from non-relevant ones, which every expansion method is a descendant of. Cited by name.
• A Comparative Study of Methods for Estimating Query Language Models with Pseudo Feedback — Yuanhua Lv and ChengXiang Zhai, ACM CIKM, 2009. Pseudo-relevance feedback compared across estimation methods, including when expansion hurts. Cited by name.
• Query Expansion Using Lexical-Semantic Relations — Ellen M. Voorhees, ACM SIGIR, 1994. The early and still-relevant finding that thesaurus-based expansion helps short queries and harms well-specified ones. Cited by name.

Conversational reformulation

• Can You Unpack That? Learning to Rewrite Questions-in-Context — Ahmed Elgohary, Denis Peskov and Jordan Boyd-Graber, EMNLP, 2019. Rewriting a follow-up turn into a standalone question, and the dataset that made the task measurable. Cited by name.
• TREC Conversational Assistance Track (CAsT) — the TREC track running from 2019, which defines conversational passage retrieval with resolved and unresolved query variants, and is the standard setting for measuring context resolution. Cited by name.

Generation-based reformulation

• Precise Zero-Shot Dense Retrieval without Relevance Labels (Gao, Ma, Lin & Callan, 2022), arXiv:2212.10496 — generating a hypothetical document and embedding it in place of the query, with the conditions under which it helps.
• Query2doc: Query Expansion with Large Language Models — Liang Wang, Nan Yang and Furu Wei, EMNLP, 2023. Expanding a query with a model-written pseudo-document, and its interaction with sparse and dense retrieval respectively. Cited by name.

No verified reference is recorded for constraint pinning across a rewrite, for rewrite caching keyed on resolved context, or for fallback policies when a rewrite retrieves worse than the original. The published work evaluates reformulation strategies and leaves the surrounding safety mechanics to implementers.