Weasel: Out-of-Domain Generalization for Web Agents via Importance-Diversity Data Selection

Fatemeh Pesaran zadeh, Seyeon Choi, Xing Han Lù, Siva Reddy, Gunhee Kim

Keywords: Large language models, web agent, data selection, out-of-domain generalization, training efficiency, real-world agents, efficiency and sustainability

TL;DR: We propose Weasel, a data selection method for out-of-domain generalization in Web agent training.

Abstract: Large language models (LLMs) have enabled web agents that follow natural language goals through multi-step browser interactions. However, agents fine-tuned on specific trajectories and domain often struggle to generalize out of domain, and offline training can be compute-inefficient due to noisy, redundant trajectories and long accessibility-tree (AXTree) states. To address both issues, we propose Weasel, a trajectory selection method for offline training of web agents. Weasel selects a fixed-budget subset of trajectory steps by optimizing an objective that balances unary importance with pairwise diversity over states, websites, and interaction patterns, solving efficiently with a greedy algorithm. We further improve efficiency with action-centered AXTree pruning that keeps only content around the ground-truth action target, and we mitigate style mismatch for reasoning-native models by replacing expert traces with model-generated, style-consistent rationales. Across AgentTrek and NNetNav training datasets, evaluations in WebArena, WorkArena, and MiniWob, and experiments with Qwen2.5-7B, Gemma3-4B, and Qwen3-8B, Weasel improves out-of-domain performance while reducing training cost, producing roughly 9.7-12.5$\times$ training speedups over standard fine-tuning.

Submission Number: 37