{"ID":2884135,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2508.07313","arxiv_id":"2508.07313","title":"DocR1: Evidence Page-Guided GRPO for Multi-Page Document Understanding","abstract":"Understanding multi-page documents poses a significant challenge for multimodal large language models (MLLMs), as it requires fine-grained visual comprehension and multi-hop reasoning across pages. While prior work has explored reinforcement learning (RL) for enhancing advanced reasoning in MLLMs, its application to multi-page document understanding remains underexplored. In this paper, we introduce DocR1, an MLLM trained with a novel RL framework, Evidence Page-Guided GRPO (EviGRPO). EviGRPO incorporates an evidence-aware reward mechanism that promotes a coarse-to-fine reasoning strategy, guiding the model to first retrieve relevant pages before generating answers. This training paradigm enables us to build high-quality models with limited supervision. To support this, we design a two-stage annotation pipeline and a curriculum learning strategy, based on which we construct two datasets: EviBench, a high-quality training set with 4.8k examples, and ArxivFullQA, an evaluation benchmark with 8.6k QA pairs based on scientific papers. Extensive experiments across a wide range of benchmarks demonstrate that DocR1 achieves state-of-the-art performance on multi-page tasks, while consistently maintaining strong results on single-page benchmarks.","short_abstract":"Understanding multi-page documents poses a significant challenge for multimodal large language models (MLLMs), as it requires fine-grained visual comprehension and multi-hop reasoning across pages. While prior work has explored reinforcement learning (RL) for enhancing advanced reasoning in MLLMs, its application to mu...","url_abs":"https://arxiv.org/abs/2508.07313","url_pdf":"https://arxiv.org/pdf/2508.07313v3","authors":"[\"Junyu Xiong\",\"Yonghui Wang\",\"Weichao Zhao\",\"Chenyu Liu\",\"Bing Yin\",\"Wengang Zhou\",\"Houqiang Li\"]","published":"2025-08-10T12:03:45Z","proceeding":"cs.CV","tasks":"[\"cs.CV\"]","methods":"[\"Reinforcement Learning\",\"Large Language Model\",\"Language Model\"]","has_code":false}