DocDancer: Towards Agentic Document-Grounded Information Seeking

TL;DR

This paper introduces DocDancer, an open-source agent framework for document question answering that explicitly models exploration and comprehension, trained with synthetic data to improve long-context document understanding.

Contribution

It presents a novel tool-driven agent framework for DocQA, along with an end-to-end training pipeline using synthetic data, addressing data scarcity and enhancing long-context understanding.

Findings

Effective on long-context document benchmarks

Outperforms existing models in document comprehension tasks

Provides insights into agentic tool design and synthetic data use

Abstract

Document Question Answering (DocQA) focuses on answering questions grounded in given documents, yet existing DocQA agents lack effective tool utilization and largely rely on closed-source models. In this work, we introduce DocDancer, an end-to-end trained open-source Doc agent. We formulate DocQA as an information-seeking problem and propose a tool-driven agent framework that explicitly models document exploration and comprehension. To enable end-to-end training of such agents, we introduce an Exploration-then-Synthesis data synthesis pipeline that addresses the scarcity of high-quality training data for DocQA. Training on the synthesized data, the trained models on two long-context document understanding benchmarks, MMLongBench-Doc and DocBench, show their effectiveness. Further analysis provides valuable insights for the agentic tool design and synthetic data.