Entry

Modeling Distinct Human Interaction in Web Agents

Faria Huq, Zora Zhiruo Wang, Zhanqiu Guo, Venu Arvind Arangarajan, Tianyue Ou, Frank Xu, Shuyan Zhou, Graham Neubig, Jeffrey P. Bigham

Synopsis

Introduces CowCorpus (400 user trajectories with 4,200+ interactions) and four interaction patterns; modeling distinct interaction styles yields 26.5% improvement in user-rated agent usefulness.

Keywords

·web agents ·intervention modeling ·CowCorpus ·human-in-the-loop

Open paper ↗ arXiv ↗ Report issue ↗

Related entries

Strategic Algorithmic Monoculture: Experimental Evidence from Coordination Games

April 10, 2026 · arXiv
When Stereotypes GTG: The Impact of Predictive Text Suggestions on Gender Bias in Human-AI Co-Writing

September 30, 2024 · CHI 2026
Homogenizing effect of large language models (LLMs) on creative diversity: An empirical comparison of human and ChatGPT writing

September 2024 · ScienceDirect 2025
Does Writing with Language Models Reduce Content Diversity?

September 11, 2023 · ICLR 2024