Query Understanding in the Age of Large Language Models (arXiv 2306.16004)

Forum topic · 小凯 · 2026-07-05

Summary

This June 2023 arXiv paper, 'Query Understanding in the Age of Large Language Models' by Avishek Anand, Venktesh V, Abhijit Anand, and Vinay Setty (arXiv:2306.16004), surveys how large language models (LLMs) are reshaping query understanding in information retrieval and search systems. It provides a unified perspective on the query processing pipeline—including query representation, expansion, rewriting, decomposition, and intent classification—and examines how LLM-based techniques compare to traditional neural and classical IR methods. The paper covers benchmarks, evaluation protocols, and efficiency considerations, and discusses open challenges such as evaluation credibility, latency and cost, hallucination and safety, cross-lingual and multimodal generalization, and the transition from research prototypes to industrial systems with tool calling and agentic search. Relevant to researchers and engineers working on search, recommendation, RAG, and LLM-powered retrieval systems.

Query Understanding in the Age of Large Language Models (arXiv:2306.16004)

Authors: Avishek Anand, Venktesh V, Abhijit Anand, Vinay Setty
Published: June 2023 (arXiv)
Source: https://arxiv.org/abs/2306.16004
Type: Survey / academic paper
Section: Query Understanding

Overview

This paper surveys query understanding in the era of large language models (LLMs). It examines how LLMs are transforming core query processing tasks in information retrieval—such as query expansion, rewriting, intent classification, and query decomposition—and positions these advances against classical and pre-LLM neural approaches.

Key points

Provides a unified view of query understanding work, organizing previously scattered research into a comparable framework.
Decomposes the pipeline into components: query representation, retrievers, re-rankers, planners/generators, and feedback mechanisms.
Covers how LLMs enable richer natural-language interaction, multi-hop reasoning, and conversational search scenarios in open-domain and enterprise retrieval.
Discusses interfaces with emerging paradigms such as LLM tool calling, reinforcement learning, and multi-agent collaboration, and the path from research prototypes to industrial systems.
Surveys evaluation practices and benchmarks (e.g., MS MARCO, BEIR, Natural Questions) and metrics such as nDCG@10, MRR, and Recall@k, while noting the growing gap between offline metrics and online user satisfaction.

Open challenges

Credibility of LLM-based evaluation (LLM-as-judge) and the need for cross-validation with human assessment.
Latency and cost constraints of LLM inference in production search stacks.
Hallucination, safety, and robustness of retrieval-augmented generation (RAG) pipelines.
Cross-lingual and multimodal generalization; benchmark coverage skewed toward English.

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)
Few-Shot Generative Conversational Query Rewriting (SIGIR 2020, DOI: 10.1145/3397271.3401323)
Hierarchical query classification in e-commerce search (WWW 2024)
Hypothetical Documents or Knowledge Leakage? Rethinking LLM-based Query Expansion (arXiv:2504.14175)

> Note: Quantitative results should be verified against the original PDF; this entry summarizes the survey's scope and contributions based on the abstract and public metadata.