WebLists: Extracting Structured Information From Complex Interactive Websites Using Executable LLM Agents

Arth BohraCollege of Computing, Data Science, and SocietyUniversity of California BerkeleyBerkeley, CA, USA [email protected], Manvel Saroyan, Danil Melkozerov, Vahe Karufanyan Gabriel Maher, Pascal Weinberger, Artem Harutyunyan, Giovanni Campagna Bardeen, Inc.San Francisco, CA, USA {manvel,dan · 2025

Read WebLists: Extracting Structured Information From Complex Interactive Websites Using Executable LLM Agents by Arth BohraCollege of Computing, Data Science, and SocietyUniversity of California BerkeleyBerkeley, CA, USA [email protected], Manvel Saroyan, Danil Melkozerov, Vahe Karufanyan Gabriel Maher, Pascal Weinberger, Artem Harutyunyan, Giovanni Campagna Bardeen, Inc.San Francisco, CA, USA {manvel,dan (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-12682.txt.

Description

Bohra and co-authors benchmark a neglected web task: pulling complete, structured data sets out of interactive business websites. Agents shine at clicking through transactions, but extraction at scale is different: the agent must configure a page correctly and harvest every row. Their 2025 benchmark, WebLists, holds two hundred such tasks across four enterprise use cases.

Leading agents did poorly, recovering three percent of rows with search help and thirty-one percent at best, despite strong question-answering form. Their system, BardeenAgent, converts one successful run into a repeatable program, exploiting the regular structure of HTML to build selectors that catch every item. It reached sixty-six percent recall at a third of the cost per row. The authors argue that programs, not improvisation, are how agents will do data work at scale.

Catalog record

CREATORArth BohraCollege of Computing, Data Science, and SocietyUniversity of California BerkeleyBerkeley, CA, USA [email protected], Manvel Saroyan, Danil Melkozerov, Vahe Karufanyan Gabriel Maher, Pascal Weinberger, Artem Harutyunyan, Giovanni Campagna Bardeen, Inc.San Francisco, CA, USA {manvel,dan
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-12682

Browse the full library · About the Tower of Babel