RAGShaper: Eliciting Sophisticated Agentic RAG Skills via Automated Data Synthesis

TL;DR

RAGShaper tackles the data scarcity and noise inherent in training Agentic RAG systems by introducing an automated data-synthesis pipeline. A central component, InfoCurator, builds dense information trees augmented with adversarial distractors across Perception and Cognition, while a constrained teacher-guided process elicits robust, error-correcting trajectories. The framework's four-phase workflow—Information Curation, QA Synthesis, Behavior Elicitation, and Training—yields high-quality, multi-hop reasoning data, demonstrated to surpass baselines on four open-domain benchmarks and to generalize across backbones. The work offers a scalable path to robust agentic retrieval by enabling efficient, diverse, and noise-aware training data generation with broad practical impact for complex information-seeking tasks.

Abstract

Agentic Retrieval-Augmented Generation (RAG) empowers large language models to autonomously plan and retrieve information for complex problem-solving. However, the development of robust agents is hindered by the scarcity of high-quality training data that reflects the noise and complexity of real-world retrieval environments. Conventional manual annotation is unscalable and often fails to capture the dynamic reasoning strategies required to handle retrieval failures. To bridge this gap, we introduce RAGShaper, a novel data synthesis framework designed to automate the construction of RAG tasks and robust agent trajectories. RAGShaper incorporates an InfoCurator to build dense information trees enriched with adversarial distractors spanning Perception and Cognition levels. Furthermore, we propose a constrained navigation strategy that forces a teacher agent to confront these distractors, thereby eliciting trajectories that explicitly demonstrate error correction and noise rejection. Comprehensive experiments confirm that models trained on our synthesized corpus significantly outperform existing baselines, exhibiting superior robustness in noise-intensive and complex retrieval tasks.

Figures (14)

• Figure 1: Limitations of human annotation of the agentic RAG dataset, which can be mitigated by automatic synthesis by the agent curator.
• Figure 2: Overview of RAGShaper.
• Figure 3: Tool call statistics on 4.5k data on both RAGShaper and HL-Data.
• Figure 4: Trajectory analysis.
• Figure 5: An illustrative example of the distractor taxonomy used in data synthesis. The figure visualizes four distinct categories of cognitive traps (Doppelganger, Fragmented Puzzle, False Shortcut, and Subjective Fallacy) designed to challenge the agent's retrieval and reasoning robustness.