Year: 2026
Type: preprint
Abstract: Deep Research Agents have shown strong capability in multi-step information retrieval, reasoning, and long-form report generation, but existing benchmarks and systems remain predominantly text-centric, with limited evaluation of whether visual elements are factually reliable and well aligned with the surrounding analysis. To address this gap, we introduce TVIR (Text-Visual Interleaved Report Generation), which includes TVIR-Bench, a benchmark of 100 expert-curated multimodal deep research tasks that require visual elements to serve specific analytical sub-goals, and TVIR-Agent, a hierarchical multi-agent framework that serves as a strong baseline for constructing outlines, retrieving images, generating charts with traceable sources, and composing reports through context-aware sequential writing. We further develop a dual-path evaluation framework that combines Textual Assessment and Visual Assessment. Experiments across nine deep research systems show that TVIR-Agent achieves strong overall performance, underscoring the importance of explicit multimodal design and evaluation for evidence-driven report generation.
Source: arXiv (Cornell University)
Authors: Ma, Xinkai, Bai, Zhiqi, Zhang, Dingling, Liu, Pei, Yuan, Yishuo, Zhu, He, Wang, Jiakai, Xie, Qianqian, Zhao, Yifan, Yang, Xinlong, Cong, Hao, Yao, Zhiheng, Xie, Fengxia, Xu, Zihao, Xu, Haoran, Wang, Zhaohui, Liu, Minghao, Lin, Shirong, Tan, Yingshui, Xu, Yuchi, Su, Wenbo, Zhang, Zhaoxiang, Zheng, Bo, Liu, Jiaheng
Institution: Alibaba Group (United States)
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7163200102
Cites:
Cited by:
Related to: