Planner-Centric Reinforcement Learning for Deep Research with Structure-Aware Reward

DecomposeR introduces a graph-based planning framework to improve multi-step research and synthesis in LLMs.

Deep research tasks require LLMs to plan what to investigate, retrieve evidence, and synthesize long-form answers across multiple branches of inquiry. Existing training paradigms either rely on short-form verifiable QA as a proxy or optimize monolithic long trajectories, which ma...

ArXiv CS.AI2026-06-01