Branch-and-Browse: Efficient and Controllable Web Exploration with Tree-Structured Reasoning and Action Memory

2025-10-17

AI Summary

Existing LLM-based web agents struggle with multi-step, goal-directed tasks—such as information retrieval, report generation, and online transactions—in open-web environments due to insufficient reasoning depth, ineffective backtracking, and low computational efficiency. To address these challenges, we propose a Tree-based Reasoning-and-Action (TRA) architecture that integrates subtask decomposition, context-and-action memory, webpage state replay, and background-reasoning-guided exploration. This enables fine-grained, backtrackable multi-step reasoning and cross-session knowledge reuse. Evaluated on the WebArena benchmark, our approach achieves a 35.8% task success rate and reduces execution time by up to 40.4%, significantly outperforming prior methods. Our core contributions are threefold: (1) the first integration of structured tree search into web agents; (2) the introduction of shared action memory for persistent, reusable interaction history; and (3) a unified framework that jointly optimizes reasoning depth, controllability, and execution efficiency.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Multiagent Systems: Modeling other Agents
• Data Mining & Knowledge Management: Knowledge Acquisition from the Web

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Autonomous web agents powered by large language models (LLMs) show strong potential for performing goal-oriented tasks such as information retrieval, report generation, and online transactions. These agents mark a key step toward practical embodied reasoning in open web environments. However, existing approaches remain limited in reasoning depth and efficiency: vanilla linear methods fail at multi-step reasoning and lack effective backtracking, while other search strategies are coarse-grained and computationally costly. We introduce Branch-and-Browse, a fine-grained web agent framework that unifies structured reasoning-acting, contextual memory, and efficient execution. It (i) employs explicit subtask management with tree-structured exploration for controllable multi-branch reasoning, (ii) bootstraps exploration through efficient web state replay with background reasoning, and (iii) leverages a page action memory to share explored actions within and across sessions. On the WebArena benchmark, Branch-and-Browse achieves a task success rate of 35.8% and reduces execution time by up to 40.4% relative to state-of-the-art methods. These results demonstrate that Branch-and-Browse is a reliable and efficient framework for LLM-based web agents.

Problem

• Enhancing reasoning depth and efficiency for web agents
• Enabling controllable multi-branch reasoning with tree structures
• Reducing computational costs through action memory sharing

Innovation

• Tree-structured exploration for controllable multi-branch reasoning
• Efficient web state replay with background reasoning
• Page action memory sharing across sessions