Few-Shot Generative Conversational Query Rewriting (SIGIR 2020)

Forum topic · 小凯 · 2026-07-05

Summary

This forum post indexes the SIGIR 2020 paper 'Few-Shot Generative Conversational Query Rewriting', which addresses conversational query rewriting in information retrieval. Conversational search systems must resolve coreferences and omitted context across dialogue turns before issuing effective search queries, but training data for such rewriting is scarce. The paper proposes a generative, few-shot approach to rewriting conversational queries into self-contained search queries, reducing the dependence on large annotated datasets. The post situates the work within the broader landscape of query understanding, retrieval-augmented generation, and LLM-era search architecture, discussing how rewriting fits into the classic recall-rank-generate pipeline and how agentic paradigms treat retrieval policies as learnable decisions. It also outlines typical evaluation setups (datasets such as MS MARCO and Natural Questions, metrics like nDCG@10 and MRR), engineering deployment concerns (latency budgets, safety, cost control), and open problems including evaluation reliability, hallucination, and cross-lingual generalization. The entry includes cross-references to related query expansion and query decomposition papers, a glossary of IR terms, and action recommendations for researchers, engineers, and product managers. Quantitative results should be verified against the original PDF at https://dl.acm.org/doi/abs/10.1145/3397271.3401323.

Few-Shot Generative Conversational Query Rewriting (SIGIR 2020)

Overview

This entry indexes the SIGIR 2020 paper "Few-Shot Generative Conversational Query Rewriting", categorized under Query Understanding.

Source: https://dl.acm.org/doi/abs/10.1145/3397271.3401323
Venue: SIGIR 2020
Type: Academic paper

Background and Motivation

Conversational search systems face persistent challenges around efficiency, scalability, and user intent understanding. Users issue follow-up queries with coreferences and omitted context (e.g., "how old is he?"), which cannot be matched effectively against an index without rewriting. Traditional pipelines separate retrieval, ranking, and generation, making it hard to meet modern expectations for natural-language interaction and multi-turn coherence. This paper proposes a generative approach to rewriting conversational queries into standalone search queries in a few-shot setting, reducing reliance on large amounts of annotated rewriting data.

Relevance to Search, Recommendation, and Personalization

The post places this work at the intersection of conversational IR and large-scale search/recommendation systems, and discusses implications that extend beyond the paper itself:

1. Architecture: Cascaded retrieval + re-ranking + generation remains mainstream, but agentic paradigms increasingly treat the *number and policy of retrieval steps* as learnable. 2. Data: High-quality instruction data and click/session logs are critical; synthetic data requires safeguards against knowledge leakage and distribution shift. 3. Evaluation: The gap between offline metrics and online satisfaction is widening; LLM-as-judge should be cross-validated with human assessment. 4. Product constraints: Latency, cost, explainability, and safety are hard requirements in industrial deployment, not just academic benchmarks.

Typical Evaluation Setup

Datasets: MS MARCO, BEIR, Natural Questions, domain-specific corpora
Metrics: nDCG@10, MRR, Recall@k, Hit@k, task success rate, latency and token cost
Baselines: BM25, dense retrieval, cross-encoder re-ranking, retrieval-free LLMs
Ablations: contributions of retrieval steps, re-ranking depth, and training data scale

*Note: specific numbers should be verified against the original PDF; this summary is based on abstract and public metadata.*

Engineering Checklist

| Item | Question | Recommendation | |------|----------|----------------| | Data | PII in training/index data? Versioning? | Partitioned indexes, anonymization, rollback-capable embedding versions | | Latency | p99 budget? Retrieval steps? | Cascades with early stopping, hot-query caching, async re-ranking | | Quality | Do offline gains translate online? | Interleaving experiments, human audits, citation checks | | Safety | Poisoning/bias from open retrieval? | Source whitelisting, adversarial detection, output filtering | | Cost | Token and GPU cost per query? | Small-model routing, distillation, hybrid sparse+dense retrieval |

Related Entries

Aligned Query Expansion: Efficient Query Expansion for Information Retrieval (arXiv 2507.11042)
Beyond the Limitation of a Single Query: Train Your LLM for Query Expansion (arXiv 2510.10009)
Decomposing Complex Queries for Tip-of-the-tongue Retrieval (arXiv 2305.15053)
Hierarchical Query Classification in E-commerce Search (WWW 2024)
Hypothetical Documents or Knowledge Leakage? Rethinking LLM-based Query Expansion (arXiv 2504.14175)
LLM-Based Query Expansion with Gaussian Kernel Semantic Enhancement

Glossary

| Term | Meaning | |------|---------| | IR | Information Retrieval | | RAG | Retrieval-Augmented Generation | | LTR | Learning to Rank | | nDCG | Normalized Discounted Cumulative Gain | | Agentic Search | Search modeled as sequential decision-making with tool calls | | Gen-IR | Generative Information Retrieval |

References

Original paper: *Few-Shot Generative Conversational Query Rewriting*, SIGIR 2020. ACM DL

Few-Shot Generative Conversational Query Rewriting (SIGIR 2020)

Summary

Few-Shot Generative Conversational Query Rewriting (SIGIR 2020)

Overview

Background and Motivation

Relevance to Search, Recommendation, and Personalization

Typical Evaluation Setup

Engineering Checklist

Related Entries

Glossary

References

Tags