Large Language Models for Information Retrieval: A Survey

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. While the neural models excel at capturing complex contextual signals and semantic nuances, they still face challenges such as data…

By the Truvace deskOctober 14, 20262 min read1 sourcePeer-reviewedTRV-2026-1332· cite & verify

In brief

This peer-reviewed survey, published in September 2025, reviews how researchers are applying large language models to information retrieval, the technology behind search engines and a component of dialogue, question-answering, and recommender systems. It traces the field from term-based methods to neural models and covers query rewriters, retrievers, rerankers, readers, and search agents.

The work matters because search tools are described as a primary means of information acquisition in daily life, so changes in how they retrieve and generate answers affect many users. The abstract says neural approaches still face data scarcity, interpretability, and inaccuracy problems, and it does not report measured performance gains or real-world deployment outcomes.

Main points

The survey covers five LLM-related components of retrieval systems: query rewriters, retrievers, rerankers, readers, and search agents.
Information retrieval has moved from term-based methods to advanced neural models.
Neural models capture complex contextual signals and semantic nuances but face data scarcity and interpretability challenges.
The authors argue modern systems need both traditional sparse retrieval with rapid response and neural architectures with strong language understanding.

The problem

Neural retrieval models still struggle with data scarcity, limited interpretability, and producing plausible but potentially inaccurate responses, so search systems need to blend them with traditional methods.

The rundown

The abstract frames information retrieval systems such as search engines as embedded in daily life and as building blocks for dialogue, question-answering, and recommender systems. That breadth explains why improvements or weaknesses in retrieval methods carry wide practical relevance.

It describes a research trajectory in which term-based sparse methods offer rapid response while neural language models offer strong language understanding. The authors say this evolution calls for combining the two rather than replacing one with the other.

The authors say recent work seeks to use LLMs to improve IR systems and that the pace of research makes consolidation necessary. The survey aims to provide an overview and nuanced insights across the main system components.