Deep Research: A Systematic Survey
Zhengliang Shi, Yiqun Chen, Haitao Li, Weiwei Sun, Shiyu Ni, Yougang Lyu, Run-Ze Fan, Bowen Jin, Yixuan Weng, Minjun Zhu, Qiujie Xie, Xinyu Guo, Qu Yang, Jiayi Wu, Jujia Zhao, Xiaqiang Tang, Xinbei Ma, Cunxiang Wang, Jiaxin Mao, Qingyao Ai, Jen-Tse Huang, Wenxuan Wang
2025-12-03
Summary
This paper is a comprehensive overview of a new approach to using large language models (LLMs) called 'Deep Research'. It explains how LLMs are moving beyond just generating text to actually *doing* research and solving complex problems by using tools like search engines.
What's the problem?
While LLMs are getting really good at understanding and creating text, they often struggle with tasks that require careful thinking, using information from multiple sources, and making sure the information they provide is accurate and verifiable. Simply asking an LLM a question (single-shot prompting) or giving it some relevant text to work with (retrieval-augmented generation) isn't always enough for these more demanding tasks.
What's the solution?
The paper breaks down 'Deep Research' into a three-step process and identifies four key parts that make it work: figuring out what questions to ask, finding the information needed, keeping track of what's been found, and then putting it all together into an answer. It also discusses different ways to improve these systems, like carefully crafting prompts, training the LLM with specific examples, and using reinforcement learning to help the LLM learn to research more effectively.
Why it matters?
This research is important because it shows how we can build LLMs that aren't just good at talking *about* things, but can actually *do* things like research, analyze information, and solve complex problems. This opens up possibilities for using LLMs in fields like science, journalism, and any area where reliable information and critical thinking are essential. The paper also provides a guide for future researchers working in this area.
Abstract
Large language models (LLMs) have rapidly evolved from text generators into powerful problem solvers. Yet, many open tasks demand critical thinking, multi-source, and verifiable outputs, which are beyond single-shot prompting or standard retrieval-augmented generation. Recently, numerous studies have explored Deep Research (DR), which aims to combine the reasoning capabilities of LLMs with external tools, such as search engines, thereby empowering LLMs to act as research agents capable of completing complex, open-ended tasks. This survey presents a comprehensive and systematic overview of deep research systems, including a clear roadmap, foundational components, practical implementation techniques, important challenges, and future directions. Specifically, our main contributions are as follows: (i) we formalize a three-stage roadmap and distinguish deep research from related paradigms; (ii) we introduce four key components: query planning, information acquisition, memory management, and answer generation, each paired with fine-grained sub-taxonomies; (iii) we summarize optimization techniques, including prompting, supervised fine-tuning, and agentic reinforcement learning; and (iv) we consolidate evaluation criteria and open challenges, aiming to guide and facilitate future development. As the field of deep research continues to evolve rapidly, we are committed to continuously updating this survey to reflect the latest progress in this area.