Large Language Models for Information Retrieval: A Survey (arXiv 2308.07107)

Forum topic · 小凯 · 2026-07-05

Summary

This survey examines how large language models (LLMs) such as ChatGPT and GPT-4 can enhance information retrieval (IR) systems. Written by Yutao Zhu, Huaying Yuan, Shuting Wang, and colleagues (10 authors), it was released on arXiv in August 2023. The paper traces IR's evolution from term-based sparse methods through neural dense retrieval to LLM-powered architectures, and systematically reviews how LLMs improve four key pipeline components: query rewriters, retrievers, rerankers, and readers. It argues that traditional methods (fast sparse retrieval) and neural models (strong semantic understanding) are complementary, and that LLMs help address challenges like data scarcity and interpretability while introducing new risks such as hallucination. The survey also identifies emerging directions including search agents, retrieval-augmented generation (RAG), and tool use, along with open problems in evaluation, latency, cost, safety, and multilingual/multimodal coverage. It serves as a foundational reference for researchers and engineers working on LLM-based search, covering benchmarks like MS MARCO and BEIR, metrics such as nDCG and MRR, and comparisons between dense retrieval, late interaction (e.g., ColBERT), generative IR, and agentic search paradigms.

Large Language Models for Information Retrieval: A Survey

This post introduces and annotates the survey "Large Language Models for Information Retrieval: A Survey" (arXiv:2308.07107), published August 14, 2023, by Yutao Zhu, Huaying Yuan, Shuting Wang, Jiongnan Liu, Wenhan Liu, Chenlong Deng, and colleagues (10 authors total).

Overview

The survey investigates the confluence of LLMs and IR systems. As the authors state in their abstract:

> As a primary means of information acquisition, information retrieval (IR) systems, such as search engines, have integrated themselves into our daily lives. These systems also serve as components of dialogue, question-answering, and recommender systems. The trajectory of IR has evolved dynamically from its origins in term-based methods to its integration with advanced neural models.

Neural models excel at capturing contextual signals and semantic nuances but still face data scarcity, interpretability, and hallucination challenges. The paper argues for combining traditional methods (fast term-based sparse retrieval) with modern neural architectures (powerful language understanding), and reviews how LLMs improve IR across query rewriters, retrievers, rerankers, and readers, plus emerging directions such as search agents.

Key points

Provides a unified perspective organizing scattered LLM-for-IR work into a comparable framework.
Decomposes the IR pipeline into modular components: representation learning, retrievers, rerankers, planners, generators, and feedback mechanisms.
Covers four main technical lines: dense retrieval (high recall, low latency), late interaction (e.g., ColBERT — higher precision but larger indexes), generative IR (directly generating docids, simplifying cascades), and agentic search (search as sequential decision-making with multi-hop and self-reflection).
Reviews LLM integration patterns: RAG, search agents, and tool/API use — flexible and interpretable, but with latency and error-propagation tradeoffs.
Identifies open problems: evaluation trustworthiness, latency and cost, hallucination and safety, cross-lingual and multimodal expansion.

Research timeline

2019–2021: BERT reranking and DPR establish neural retrieval foundations.
2022–2023: RAG and FreshLLM drive retrieval-generation fusion.
2024+: Conversational/agentic search and generative recommenders (Gen-RecSys) surge.
2025–2026: RL-trained search agents, Deep Research, and GraphRAG become new growth areas.

Evaluation paradigms covered

Datasets: MS MARCO, BEIR, Natural Questions, domain-specific corpora.
Metrics: nDCG@10, MRR, Recall@k, Hit@k, human preference, task success rate, latency, token cost.
Baselines: BM25, dense retrieval, cross-encoder reranking, retrieval-free LLMs, commercial search APIs.
Ablations: contribution of retrieval steps, reranking depth, and training data scale.

Insights for practitioners

1. Architecture: cascaded retrieve → rerank → generate remains mainstream, but agentic paradigms make retrieval strategy itself learnable. 2. Data: high-quality instruction data and click/session logs matter; synthetic data requires leakage and distribution-shift safeguards. 3. Evaluation: the offline-online gap is widening; LLM-as-judge needs cross-validation with human assessment.

Related entries

A Comprehensive Survey on Reinforcement Learning-based Agentic Search
A Survey of Conversational Search
A Survey of LLM Empowered Agents for Recommendation
A Survey of Model Architectures in Information Retrieval
A Survey on Knowledge-Oriented Retrieval-Augmented Generation

Source: https://arxiv.org/abs/2308.07107