WebDART: Dynamic Decomposition and Re-planning for Complex Web Tasks

2025-10-07

AI Summary

Existing large language model (LLM)-based agents perform well on simple web tasks but struggle with complex ones requiring long-horizon navigation, large-scale information extraction, and constraint-aware reasoning. To address this, we propose a dynamic task decomposition and real-time replanning framework: the target task is adaptively decomposed into three subtask types—navigation, information extraction, and action execution—and context-triggered replanning is performed upon page loading, leveraging newly observed cues (e.g., filtering options or structural shortcuts) to refine subsequent steps and eliminate redundant actions. Our approach integrates stepwise reasoning, context-aware subtask partitioning, and a lightweight replanning algorithm, significantly enhancing robustness and efficiency for long-horizon tasks. On WebChoreArena, our method improves task success rate by 13.7 percentage points; on WebArena, it maintains state-of-the-art performance while reducing average navigation steps by 14.7.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Multiagent Systems: Multiagent Planning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large language model (LLM) agents are becoming competent at straightforward web tasks, such as opening an item page or submitting a form, but still struggle with objectives that require long horizon navigation, large scale information extraction, and reasoning under constraints. We present WebDART, a general framework that enables a single LLM to handle such complex chores. WebDART (i) dynamically decomposes each objective into three focused subtasks: navigation, information extraction, and execution, so the model concentrates on one skill at a time, and (ii) continuously replans the decomposition as new webpages are revealed, taking advantage of newly discovered filters or shortcuts and avoiding redundant exploration. Evaluated on WebChoreArena, WebDART lifts success rates by up to 13.7 percentage points over previous SOTA agents, while matching their performance on the easier WebArena suite and completing tasks with up to 14.7 fewer navigation steps.

Problem

• Handles complex web tasks requiring long horizon navigation
• Dynamically decomposes objectives into focused subtasks for execution
• Improves success rates while reducing redundant exploration steps

Innovation

• Dynamically decomposes objectives into three subtasks
• Continuously replans decomposition as new webpages appear
• Enhances success rates and reduces navigation steps