Guardado en:
Detalles Bibliográficos
Autores principales: Huang, Yuxuan, Chen, Yihang, He, Zhiyuan, Chen, Yuxiang, Lee, Ka Yiu, Zhou, Huichi, Luo, Weilin, Fang, Meng, Wang, Jun
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2604.27221
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911633575510016
author Huang, Yuxuan
Chen, Yihang
He, Zhiyuan
Chen, Yuxiang
Lee, Ka Yiu
Zhou, Huichi
Luo, Weilin
Fang, Meng
Wang, Jun
author_facet Huang, Yuxuan
Chen, Yihang
He, Zhiyuan
Chen, Yuxiang
Lee, Ka Yiu
Zhou, Huichi
Luo, Weilin
Fang, Meng
Wang, Jun
contents Agentic web search increasingly faces two distinct demands: deep reasoning over a single target, and structured aggregation across many entities and heterogeneous sources. Current systems struggle on both fronts. Breadth-oriented tasks demand schema-aligned outputs with wide coverage and cross-entity consistency, while depth-oriented tasks require coherent reasoning over long, branching search trajectories. We introduce \textbf{Web2BigTable}, a multi-agent framework for web-to-table search that supports both regimes. Web2BigTable adopts a bi-level architecture in which an upper-level orchestrator decomposes the task into sub-problems and lower-level worker agents solve them in parallel. Through a closed-loop run--verify--reflect process, the framework jointly improves decomposition and execution over time via persistent, human-readable external memory, with self-evolving updates to each single-agent. During execution, workers coordinate through a shared workspace that makes partial findings visible, allowing them to reduce redundant exploration, reconcile conflicting evidence, and adapt to emerging coverage gaps. Web2BigTable sets a new state of the art on WideSearch, reaching an Avg@4 Success Rate of \textbf{38.50} ($7.5\times$ the second best at 5.10), Row F1 of \textbf{63.53} (+25.03 over the second best), and Item F1 of \textbf{80.12} (+14.42 over the second best). It also generalises to depth-oriented search on XBench-DeepSearch, achieving 73.0 accuracy. Code is available at https://github.com/web2bigtable/web2bigtable.
format Preprint
id arxiv_https___arxiv_org_abs_2604_27221
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Web2BigTable: A Bi-Level Multi-Agent LLM System for Internet-Scale Information Search and Extraction
Huang, Yuxuan
Chen, Yihang
He, Zhiyuan
Chen, Yuxiang
Lee, Ka Yiu
Zhou, Huichi
Luo, Weilin
Fang, Meng
Wang, Jun
Artificial Intelligence
Agentic web search increasingly faces two distinct demands: deep reasoning over a single target, and structured aggregation across many entities and heterogeneous sources. Current systems struggle on both fronts. Breadth-oriented tasks demand schema-aligned outputs with wide coverage and cross-entity consistency, while depth-oriented tasks require coherent reasoning over long, branching search trajectories. We introduce \textbf{Web2BigTable}, a multi-agent framework for web-to-table search that supports both regimes. Web2BigTable adopts a bi-level architecture in which an upper-level orchestrator decomposes the task into sub-problems and lower-level worker agents solve them in parallel. Through a closed-loop run--verify--reflect process, the framework jointly improves decomposition and execution over time via persistent, human-readable external memory, with self-evolving updates to each single-agent. During execution, workers coordinate through a shared workspace that makes partial findings visible, allowing them to reduce redundant exploration, reconcile conflicting evidence, and adapt to emerging coverage gaps. Web2BigTable sets a new state of the art on WideSearch, reaching an Avg@4 Success Rate of \textbf{38.50} ($7.5\times$ the second best at 5.10), Row F1 of \textbf{63.53} (+25.03 over the second best), and Item F1 of \textbf{80.12} (+14.42 over the second best). It also generalises to depth-oriented search on XBench-DeepSearch, achieving 73.0 accuracy. Code is available at https://github.com/web2bigtable/web2bigtable.
title Web2BigTable: A Bi-Level Multi-Agent LLM System for Internet-Scale Information Search and Extraction
topic Artificial Intelligence
url https://arxiv.org/abs/2604.27221