Web-Shepherd: Advancing PRMs for Reinforcing Web Agents

• URL: http://arxiv.org/abs/2505.15277v1
• Date: Wed, 21 May 2025 08:56:55 GMT
• Title: Web-Shepherd: Advancing PRMs for Reinforcing Web Agents
• Authors: Hyungjoo Chae, Sunghwan Kim, Junhee Cho, Seungone Kim, Seungjun Moon, Gyeom Hwangbo, Dongha Lim, Minjin Kim, Yeonjun Hwang, Minju Gwak, Dongwook Choi, Minseok Kang, Gwanhoon Im, ByeongUng Cho, Hyojun Kim, Jun Hee Han, Taeyoon Kwon, Minju Kim, Beong-woo Kwak, Dongjin Kang, Jinyoung Yeo,
• Abstract summary: We propose the first process reward model (PRM) called Web-Shepherd to assess web navigation trajectories in a step-level.<n>In experiments, we observe that our Web-Shepherd achieves about 30 points better accuracy compared to using GPT-4o on WebRewardBench.
• Score: 12.928605558358464
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Web navigation is a unique domain that can automate many repetitive real-life tasks and is challenging as it requires long-horizon sequential decision making beyond typical multimodal large language model (MLLM) tasks. Yet, specialized reward models for web navigation that can be utilized during both training and test-time have been absent until now. Despite the importance of speed and cost-effectiveness, prior works have utilized MLLMs as reward models, which poses significant constraints for real-world deployment. To address this, in this work, we propose the first process reward model (PRM) called Web-Shepherd which could assess web navigation trajectories in a step-level. To achieve this, we first construct the WebPRM Collection, a large-scale dataset with 40K step-level preference pairs and annotated checklists spanning diverse domains and difficulty levels. Next, we also introduce the WebRewardBench, the first meta-evaluation benchmark for evaluating PRMs. In our experiments, we observe that our Web-Shepherd achieves about 30 points better accuracy compared to using GPT-4o on WebRewardBench. Furthermore, when testing on WebArena-lite by using GPT-4o-mini as the policy and Web-Shepherd as the verifier, we achieve 10.9 points better performance, in 10 less cost compared to using GPT-4o-mini as the verifier. Our model, dataset, and code are publicly available at LINK.

Related papers

• WebWalker: Benchmarking LLMs in Web Traversal [64.48425443951749]
  We introduce WebWalkerQA, a benchmark designed to assess the ability of LLMs to perform web traversal.<n>We propose WebWalker, which is a multi-agent framework that mimics human-like web navigation through an explore-critic paradigm.
  arXiv  Detail & Related papers  (2025-01-13T18:58:07Z)
• AgentOccam: A Simple Yet Strong Baseline for LLM-Based Web Agents [52.13695464678006]
  This study enhances an LLM-based web agent by simply refining its observation and action space.<n>AgentOccam surpasses the previous state-of-the-art and concurrent work by 9.8 (+29.4%) and 5.9 (+15.8%) absolute points respectively.
  arXiv  Detail & Related papers  (2024-10-17T17:50:38Z)
• AutoWebGLM: A Large Language Model-based Web Navigating Agent [33.55199326570078]
  We develop the open AutoWebGLM based on ChatGLM3-6B. Inspired by human browsing patterns, we first design an HTML simplification algorithm to represent webpages. We then employ a hybrid human-AI method to build web browsing data for curriculum training.
  arXiv  Detail & Related papers  (2024-04-04T17:58:40Z)
• Design2Code: Benchmarking Multimodal Code Generation for Automated Front-End Engineering [74.99736967448423]
  We construct Design2Code - the first real-world benchmark for this task.<n>We manually curate 484 diverse real-world webpages as test cases and develop a set of automatic evaluation metrics.<n>Our fine-grained break-down metrics indicate that models mostly lag in recalling visual elements from the input webpages and generating correct layout designs.
  arXiv  Detail & Related papers  (2024-03-05T17:56:27Z)
• WebVoyager: Building an End-to-End Web Agent with Large Multimodal Models [65.18602126334716]
  Existing web agents typically only handle one input modality and are evaluated only in simplified web simulators or static web snapshots. We introduce WebVoyager, an innovative Large Multimodal Model (LMM) powered web agent that can complete user instructions end-to-end by interacting with real-world websites. We show that WebVoyager achieves a 59.1% task success rate on our benchmark, significantly surpassing the performance of both GPT-4 (All Tools) and the WebVoyager (text-only) setups.
  arXiv  Detail & Related papers  (2024-01-25T03:33:18Z)
• GPT-4V(ision) is a Generalist Web Agent, if Grounded [20.940613419944015]
  We show that GPT-4V can successfully complete 51.1 of the tasks on live websites if we manually ground its textual plans into actions on the websites. We propose SEEACT, a web agent that harnesses the power of LMMs for integrated visual understanding and acting on the web.
  arXiv  Detail & Related papers  (2024-01-03T08:33:09Z)
• AllTogether: Investigating the Efficacy of Spliced Prompt for Web Navigation using Large Language Models [2.234037966956278]
  We introduce AllTogether, a standardized prompt template that enhances task context representation. We evaluate the efficacy of this approach through prompt learning and instruction finetuning based on open-source Llama-2 and API-accessible GPT models.
  arXiv  Detail & Related papers  (2023-10-20T11:10:14Z)
• A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis [69.15016747150868]
  We introduce WebAgent, an agent that learns from self-experience to complete tasks on real websites. WebAgent plans ahead by decomposing instructions into canonical sub-instructions, summarizes long HTML documents into task-relevant snippets, and acts on websites. We empirically demonstrate that our modular recipe improves the success on real websites by over 50%, and that HTML-T5 is the best model to solve various HTML understanding tasks.
  arXiv  Detail & Related papers  (2023-07-24T14:56:30Z)
• WebGLM: Towards An Efficient Web-Enhanced Question Answering System with Human Preferences [32.70333236055738]
  WebGLM is a web-enhanced question-answering system based on the General Language Model (GLM) We develop WebGLM with strategies for the LLM-augmented retriever, bootstrapped generator, and human preference-aware scorer.
  arXiv  Detail & Related papers  (2023-06-13T16:57:53Z)
• Multimodal Web Navigation with Instruction-Finetuned Foundation Models [99.14209521903854]
  We study data-driven offline training for web agents with vision-language foundation models. We propose an instruction-following multimodal agent, WebGUM, that observes both webpage screenshots and HTML pages. We empirically demonstrate this recipe improves the agent's ability of grounded multimodal perception, HTML comprehension, and multi-step reasoning.
  arXiv  Detail & Related papers  (2023-05-19T17:44:34Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.