Go-Browse: Training Web Agents with Structured Exploration

2025-06-04

AI Summary

Web agents frequently fail to navigate unfamiliar websites due to insufficient environmental understanding and ineffective path planning. To address this, we propose a graph-search-driven structured exploration framework that formally models web navigation as a graph search process over reusable information—enabling cross-session knowledge sharing and scalable generation of high-quality navigation trajectories. Our method integrates graph search algorithms, tight coupling with the WebArena benchmark, fine-tuning of a 7B-language-model, and interactive, URL-grounded data collection. This yields a high-fidelity dataset comprising 10K successful trajectories and 40K interaction steps. Evaluated on WebArena, the fine-tuned model achieves a 21.7% task success rate—outperforming GPT-4o mini by 2.4 percentage points and establishing a new state-of-the-art for models of comparable scale.

Technology Category

• Search and Optimization: Learning to Search
• Planning, Routing, and Scheduling: Planning with Language Models
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web search models and ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

One of the fundamental problems in digital agents is their lack of understanding of their environment. For instance, a web browsing agent may get lost in unfamiliar websites, uncertain what pages must be visited to achieve its goals. To address this, we propose Go-Browse, a method for automatically collecting diverse and realistic web agent data at scale through structured exploration of web environments. Go-Browse achieves efficient exploration by framing data collection as a graph search, enabling reuse of information across exploration episodes. We instantiate our method on the WebArena benchmark, collecting a dataset of 10K successful task-solving trajectories and 40K interaction steps across 100 URLs. Fine-tuning a 7B parameter language model on this dataset achieves a success rate of 21.7% on the WebArena benchmark, beating GPT-4o mini by 2.4% and exceeding current state-of-the-art results for sub-10B parameter models by 2.9%.

Problem

• Lack of web agents' understanding of their environment
• Difficulty in navigating unfamiliar websites efficiently
• Need for scalable data collection for web agent training

Innovation

• Structured exploration for web agent training
• Graph search for efficient data collection
• Fine-tuning language models with collected trajectories