WEBCOT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback

Minda Hu; Tianqing Fang; Jianshu Zhang; Junyu Ma; Zhisong Zhang; Jingyan Zhou; Hongming Zhang; Haitao Mi; Dong Yu; Irwin King

doi:10.18653/v1/2025.findings-emnlp.276

WEBCOT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback

Minda Hu (Co-first Author)
, Tianqing Fang (Co-first Author)
, Jianshu Zhang
, Junyu Ma
, Zhisong Zhang
, Jingyan Zhou
, Hongming Zhang
, Haitao Mi
, Dong Yu
, Irwin King

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

1 Citation (Scopus)

Abstract

Web agents powered by Large Language Models (LLMs) show promise for next-generation AI, but their limited reasoning in uncertain, dynamic web environments hinders robust deployment. In this paper, we identify key reasoning skills essential for effective web agents, i.e., reflection & lookahead, branching, and rollback, and curate trajectory data that exemplifies these abilities by reconstructing the agent’s (inference-time) reasoning algorithms into chain-of-thought rationales. We conduct experiments in the agent self-improving benchmark, OpenWebVoyager, and demonstrate that distilling salient reasoning patterns into the backbone LLM via simple fine-tuning can substantially enhance its performance. Our approach yields significant improvements across multiple benchmarks, including WebVoyager, Mind2web-live, and SimpleQA (web search), highlighting the potential of targeted reasoning skill enhancement for web agents. ©2025 Association for Computational Linguistics.

Original language	English
Title of host publication	Findings of the Association for Computational Linguistics
Subtitle of host publication	EMNLP 2025
Editors	Christos Christodoulopoulos, Tanmoy Chakraborty, Carolyn Rose, Violet Peng
Publisher	Association for Computational Linguistics
Pages	5155-5173
Number of pages	19
ISBN (Print)	979-8-89176-335-7
DOIs	https://doi.org/10.18653/v1/2025.findings-emnlp.276
Publication status	Published - Nov 2025
Externally published	Yes
Event	30th Conference on Empirical Methods in Natural Language Processing (EMNLP 2025) - Suzhou, China Duration: 4 Nov 2025 → 9 Nov 2025 https://aclanthology.org/volumes/2025.emnlp-main/

Publication series

Name	EMNLP - Conference on Empirical Methods in Natural Language Processing, Findings of EMNLP

Conference

Conference	30th Conference on Empirical Methods in Natural Language Processing (EMNLP 2025)
Abbreviated title	30th EMNLP
Place	China
City	Suzhou
Period	4/11/25 → 9/11/25
Internet address	https://aclanthology.org/volumes/2025.emnlp-main/

Funding

Two authors (i.e., Minda Hu, Irwin King) of the work described in this paper were partially supported by the Research Grants Council of the Hong Kong Special Administrative Region, China (CUHK 2410072, RGC R1015-23). As the first author, I would like to express my heartfelt gratitude to my family, co-authors, and advisor, Prof. Irwin King, for their unwavering support and invaluable guidance throughout this work.

Publisher's Copyright Statement

This full text is made available under CC-BY 4.0. https://creativecommons.org/licenses/by/4.0/

RGC Funding Information

RGC-funded

Access Output

10.18653/v1/2025.findings-emnlp.276

477020085.pdfFinal Published version, 667 KBLicence: CC BY

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

Hu, M., Fang, T., Zhang, J., Ma, J., Zhang, Z., Zhou, J., Zhang, H., Mi, H., Yu, D., & King, I. (2025). WEBCOT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback. In C. Christodoulopoulos, T. Chakraborty, C. Rose, & V. Peng (Eds.), Findings of the Association for Computational Linguistics: EMNLP 2025 (pp. 5155-5173). (EMNLP - Conference on Empirical Methods in Natural Language Processing, Findings of EMNLP). Association for Computational Linguistics. https://doi.org/10.18653/v1/2025.findings-emnlp.276

Hu, Minda ; Fang, Tianqing ; Zhang, Jianshu et al. / WEBCOT : Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback. Findings of the Association for Computational Linguistics: EMNLP 2025. editor / Christos Christodoulopoulos ; Tanmoy Chakraborty ; Carolyn Rose ; Violet Peng. Association for Computational Linguistics, 2025. pp. 5155-5173 (EMNLP - Conference on Empirical Methods in Natural Language Processing, Findings of EMNLP).

@inproceedings{98ba2c1f4e4446dba7396d4c0bf799fe,

title = "WEBCOT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback",

abstract = "Web agents powered by Large Language Models (LLMs) show promise for next-generation AI, but their limited reasoning in uncertain, dynamic web environments hinders robust deployment. In this paper, we identify key reasoning skills essential for effective web agents, i.e., reflection \& lookahead, branching, and rollback, and curate trajectory data that exemplifies these abilities by reconstructing the agent{\textquoteright}s (inference-time) reasoning algorithms into chain-of-thought rationales. We conduct experiments in the agent self-improving benchmark, OpenWebVoyager, and demonstrate that distilling salient reasoning patterns into the backbone LLM via simple fine-tuning can substantially enhance its performance. Our approach yields significant improvements across multiple benchmarks, including WebVoyager, Mind2web-live, and SimpleQA (web search), highlighting the potential of targeted reasoning skill enhancement for web agents. {\textcopyright}2025 Association for Computational Linguistics.",

author = "Minda Hu and Tianqing Fang and Jianshu Zhang and Junyu Ma and Zhisong Zhang and Jingyan Zhou and Hongming Zhang and Haitao Mi and Dong Yu and Irwin King",

year = "2025",

month = nov,

doi = "10.18653/v1/2025.findings-emnlp.276",

language = "English",

isbn = "979-8-89176-335-7",

series = "EMNLP - Conference on Empirical Methods in Natural Language Processing, Findings of EMNLP",

publisher = "Association for Computational Linguistics",

pages = "5155--5173",

editor = "Christodoulopoulos, \{Christos \} and Chakraborty, \{Tanmoy \} and Rose, \{Carolyn \} and Peng, \{Violet \}",

booktitle = "Findings of the Association for Computational Linguistics",

address = "United States",

note = "30th Conference on Empirical Methods in Natural Language Processing (EMNLP 2025), 30th EMNLP ; Conference date: 04-11-2025 Through 09-11-2025",

url = "https://aclanthology.org/volumes/2025.emnlp-main/",

}

Hu, M, Fang, T, Zhang, J, Ma, J, Zhang, Z, Zhou, J, Zhang, H, Mi, H, Yu, D & King, I 2025, WEBCOT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback. in C Christodoulopoulos, T Chakraborty, C Rose & V Peng (eds), Findings of the Association for Computational Linguistics: EMNLP 2025. EMNLP - Conference on Empirical Methods in Natural Language Processing, Findings of EMNLP, Association for Computational Linguistics, pp. 5155-5173, 30th Conference on Empirical Methods in Natural Language Processing (EMNLP 2025), Suzhou, China, 4/11/25. https://doi.org/10.18653/v1/2025.findings-emnlp.276

WEBCOT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback. / Hu, Minda (Co-first Author); Fang, Tianqing (Co-first Author); Zhang, Jianshu et al.
Findings of the Association for Computational Linguistics: EMNLP 2025. ed. / Christos Christodoulopoulos; Tanmoy Chakraborty; Carolyn Rose; Violet Peng. Association for Computational Linguistics, 2025. p. 5155-5173 (EMNLP - Conference on Empirical Methods in Natural Language Processing, Findings of EMNLP).

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

TY - GEN

T1 - WEBCOT

T2 - 30th Conference on Empirical Methods in Natural Language Processing (EMNLP 2025)

AU - Hu, Minda

AU - Fang, Tianqing

AU - Zhang, Jianshu

AU - Ma, Junyu

AU - Zhang, Zhisong

AU - Zhou, Jingyan

AU - Zhang, Hongming

AU - Mi, Haitao

AU - Yu, Dong

AU - King, Irwin

PY - 2025/11

Y1 - 2025/11

N2 - Web agents powered by Large Language Models (LLMs) show promise for next-generation AI, but their limited reasoning in uncertain, dynamic web environments hinders robust deployment. In this paper, we identify key reasoning skills essential for effective web agents, i.e., reflection & lookahead, branching, and rollback, and curate trajectory data that exemplifies these abilities by reconstructing the agent’s (inference-time) reasoning algorithms into chain-of-thought rationales. We conduct experiments in the agent self-improving benchmark, OpenWebVoyager, and demonstrate that distilling salient reasoning patterns into the backbone LLM via simple fine-tuning can substantially enhance its performance. Our approach yields significant improvements across multiple benchmarks, including WebVoyager, Mind2web-live, and SimpleQA (web search), highlighting the potential of targeted reasoning skill enhancement for web agents. ©2025 Association for Computational Linguistics.

AB - Web agents powered by Large Language Models (LLMs) show promise for next-generation AI, but their limited reasoning in uncertain, dynamic web environments hinders robust deployment. In this paper, we identify key reasoning skills essential for effective web agents, i.e., reflection & lookahead, branching, and rollback, and curate trajectory data that exemplifies these abilities by reconstructing the agent’s (inference-time) reasoning algorithms into chain-of-thought rationales. We conduct experiments in the agent self-improving benchmark, OpenWebVoyager, and demonstrate that distilling salient reasoning patterns into the backbone LLM via simple fine-tuning can substantially enhance its performance. Our approach yields significant improvements across multiple benchmarks, including WebVoyager, Mind2web-live, and SimpleQA (web search), highlighting the potential of targeted reasoning skill enhancement for web agents. ©2025 Association for Computational Linguistics.

UR - https://www.scopus.com/pages/publications/105028978938

UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-105028978938&origin=recordpage

U2 - 10.18653/v1/2025.findings-emnlp.276

DO - 10.18653/v1/2025.findings-emnlp.276

M3 - RGC 32 - Refereed conference paper (with host publication)

SN - 979-8-89176-335-7

T3 - EMNLP - Conference on Empirical Methods in Natural Language Processing, Findings of EMNLP

SP - 5155

EP - 5173

BT - Findings of the Association for Computational Linguistics

A2 - Christodoulopoulos, Christos

A2 - Chakraborty, Tanmoy

A2 - Rose, Carolyn

A2 - Peng, Violet

PB - Association for Computational Linguistics

Y2 - 4 November 2025 through 9 November 2025

ER -

Hu M, Fang T, Zhang J, Ma J, Zhang Z, Zhou J et al. WEBCOT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback. In Christodoulopoulos C, Chakraborty T, Rose C, Peng V, editors, Findings of the Association for Computational Linguistics: EMNLP 2025. Association for Computational Linguistics. 2025. p. 5155-5173. (EMNLP - Conference on Empirical Methods in Natural Language Processing, Findings of EMNLP). doi: 10.18653/v1/2025.findings-emnlp.276

WEBCOT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback

Abstract

Publication series

Conference

Funding

Publisher's Copyright Statement

RGC Funding Information

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

RIF: Integrating ChatGPT with Search Engine, Recommender System and Online Advertising to Enhance User Experience on Online Service Platforms

Cite this

WEBCOT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback

Abstract

Publication series

Conference

Funding

Publisher's Copyright Statement

RGC Funding Information

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Projects

RIF: Integrating ChatGPT with Search Engine, Recommender System and Online Advertising to Enhance User Experience on Online Service Platforms

Cite this