WebThinker: Empowering Large Reasoning Models with Deep Research Capability
Xiaoxi Li, Jiajie Jin, Guanting Dong, Hongjin Qian, Yutao Zhu, Yongkang Wu, Ji-Rong Wen, Zhicheng Dou
2025-05-01
Summary
This paper talks about WebThinker, a new AI system that helps large reasoning models become much better at solving tough problems by letting them search the internet for up-to-date information on their own.
What's the problem?
Most AI models can only use information they were trained on, so they struggle with questions that need the latest facts or require deep research, making them less useful for complex or current topics.
What's the solution?
The researchers built WebThinker to act like an independent research assistant. It uses a special web explorer to gather information from the internet as needed and is trained with reinforcement learning to choose the best answers based on what it finds.
Why it matters?
This matters because it lets AI tackle more challenging and real-world problems, like helping with research projects, answering tricky questions, or staying updated with new information, making these systems much more helpful and reliable.
Abstract
WebThinker, an autonomous deep research agent, improves LRM performance on complex tasks by integrating a Deep Web Explorer for dynamic web information gathering and using RL-based training through Direct Preference Optimization.