Deep Research: A Systematic Survey

2025-11-24

AI Summary

Large language models (LLMs) exhibit significant limitations in open-ended research tasks requiring multi-source verification and critical reasoning; single-turn prompting or standard retrieval-augmented generation (RAG) proves inadequate. Method: This work formally defines “Deep Research” as a three-stage paradigm—query planning, information acquisition, and answer generation—and proposes a four-component framework encompassing these stages plus memory management. It introduces a fine-grained taxonomy, integrates evaluation criteria and open challenges into a dynamically updatable research map, and employs prompt engineering, supervised fine-tuning, and agent-based reinforcement learning to optimize LLM–tool collaboration (e.g., with search engines). Contribution/Results: We deliver the first systematic architecture for Deep Research, establish key technical pathways—including query decomposition, iterative evidence synthesis, and stateful reasoning—and provide an open benchmark and living research map to guide future development.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

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.

Problem

• Enhancing LLMs for complex tasks requiring critical thinking and verifiable outputs
• Integrating external tools like search engines to empower LLMs as research agents
• Providing a systematic survey on deep research systems, components, and challenges

Innovation

• Integrates LLMs with external tools for complex tasks
• Uses multi-stage roadmap with four key components
• Applies optimization techniques like prompting and fine-tuning