From Context to Action: Analysis of the Impact of State Representation and Context on the Generalization of Multi-Turn Web Navigation Agents
Nalin Tiwary , Vardhan Dongre , Sanil Arun Chawla , Ashwin Lamani , Dilek Hakkani-Tür
- 🏛 Institutions
- UIUC
- 📅 Date
- October 31, 2024
- 📑 Publisher
- Open-World Agents @ NeurIPS 2024 (Poster)
- 💻 Env
- Web
- 🔑 Keywords
TLDR
This paper studies how interaction history and web-page representation affect multi-turn web-agent generalization on WebLINX. It shows that using longer but not excessive history and less aggressive truncation of page context improves out-of-distribution performance on unseen websites, categories, and geographic locations.
Related papers (24)
- Pack Light, CarryOn: Building Object-Oriented Context for Long-Horizon AgentsOctober 13, 2026 · arXiv
- Structured Interaction, Visual Localization, and Robust Execution for Complex Web Tasks: A Technical Report on the WebRetriever ChallengeSeptember 28, 2026 · arXiv
- WebLINX: Real-World Website Navigation with Multi-Turn DialogueFebruary 8, 2024 · ICML 2024
- AgentProg: Empowering Long-Horizon GUI Agents with Program-Guided Context ManagementDecember 11, 2025 · arXiv
- What Does "Now" Mean to Your Computer-Use Agent? Stale Screenshots, Page Loads, and Which Check to Run Before a ClickOctober 29, 2026 · arXiv
- InterA11y: Interactive Web Accessibility Error Detection with LLM AgentsOctober 29, 2026 · arXiv
- AgentHeal: Evaluating Goal-Preserving Resource Recovery of Web Agents under Access BarriersOctober 29, 2026 · arXiv
- WebAnything: From the Wild Web to Scalable Agent EnvironmentsOctober 27, 2026 · arXiv
- SkillLifeBench: Benchmarking Security Across the Lifecycle of Agent SkillsOctober 26, 2026 · arXiv
- Back to Basics: A Skill-Centric Approach to Generalist Web AgentsOctober 26, 2026 · arXiv
- AgentProofBench: Benchmarking Reusable Verifiers for Web AgentsOctober 25, 2026 · arXiv
- SkillShift-Bench: Benchmarking Skill Learning under Environment Evolution for Web AgentsOctober 24, 2026 · arXiv
- Seeing the Options: Typed Decisions over Visual Candidates with Reusable Option MemoryOctober 24, 2026 · arXiv
- Learning What to Ask for Tomorrow: Active Preference Acquisition for Long-Term Personalized Web AgentsOctober 22, 2026 · arXiv
- Step-Aligned Script Generation with Verification and Recovery for Web Automation TestingOctober 19, 2026 · arXiv
- LoWAgent: A Low-Resource Multilingual Benchmark for Web AgentsOctober 19, 2026 · arXiv
- SPIDES: Information-Flow Control for Browser AgentsOctober 16, 2026 · arXiv
- What It Is, Not What It Says: Securing Agentic Browsing Interactions with Web SemanticsOctober 15, 2026 · arXiv
- Why Generate a Click? Plug-in Decision Heads for Web AgentsOctober 14, 2026 · arXiv
- Teaching Agents in their Own Style: Blackbox Teacher Distillation for Web AgentsOctober 14, 2026 · arXiv
- STP-GUI: Spatially-Aware Token Pruning for Efficient Web-GUI AgentsOctober 14, 2026 · arXiv
- Towards Safe Autonomy for Computer UseOctober 13, 2026 · arXiv
- MHVF: Training-Free GUI Grounding via Verification-Guided Multi-Hypothesis SearchOctober 13, 2026 · arXiv
- Learning How to Remember: Memory Modality Selection for GUI AgentsOctober 13, 2026 · arXiv