Saved in:
Bibliographic Details
Main Authors: Zhang, Qintong, Lv, Xinjie, Wu, Jialong, Li, Baixuan, Tao, Zhengwei, Yan, Guochen, Zhang, Huanyao, Wang, Bin, Xu, Jiahao, Mi, Haitao, Zhang, Wentao
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.05163
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917190860537856
author Zhang, Qintong
Lv, Xinjie
Wu, Jialong
Li, Baixuan
Tao, Zhengwei
Yan, Guochen
Zhang, Huanyao
Wang, Bin
Xu, Jiahao
Mi, Haitao
Zhang, Wentao
author_facet Zhang, Qintong
Lv, Xinjie
Wu, Jialong
Li, Baixuan
Tao, Zhengwei
Yan, Guochen
Zhang, Huanyao
Wang, Bin
Xu, Jiahao
Mi, Haitao
Zhang, Wentao
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2601_05163
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DocDancer: Towards Agentic Document-Grounded Information Seeking
Zhang, Qintong
Lv, Xinjie
Wu, Jialong
Li, Baixuan
Tao, Zhengwei
Yan, Guochen
Zhang, Huanyao
Wang, Bin
Xu, Jiahao
Mi, Haitao
Zhang, Wentao
Computation and Language
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.
title DocDancer: Towards Agentic Document-Grounded Information Seeking
topic Computation and Language
url https://arxiv.org/abs/2601.05163