Research map / Multi-agent systems and games / Human-AI and goal-conditioned
End-to-End Goal-Driven Web Navigation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add recurrence (LSTM) to the small agent, (a) to (b) | NeuAgent 1x512 · WikiNav-16-4 | average reward: 2.9 -> 3.6 | Helped |
| add recurrence (LSTM) to the small agent, (a) to (b) | NeuAgent 1x512 · WikiNav-4-1 | average reward: 21.5 -> 22.0 | No clear effect |
| scale to 8 layers x 2048 LSTM units, (b) to (c) | NeuAgent-Rec, BoW query · WikiNav-16-4 | average reward: 3.6 -> 21.9 | Helped |
| scale to 8 layers x 2048 LSTM units, (b) to (c) | NeuAgent-Rec, BoW query · WikiNav-4-1 | average reward: 22.0 -> 17.7 | Hurt |
| replace BoW with attention query representation, (c) to (d) (ablation) | NeuAgent-Rec 8x2048 · WikiNav-16-4 | average reward: 21.9 -> 28.2 | Helped |
| replace BoW with attention query representation, (c) to (d) (ablation) | NeuAgent-Rec 8x2048 · WikiNav-16-1 | average reward: 8.0 -> 12.5 | Helped |
| increase query length N_q from 1 to 4 | NeuAgent (d) · WikiNav-8 | average reward: 15.8 -> 34.2 | Helped |
| increase maximum hops N_h from 4 to 16 | NeuAgent (d) · WikiNav-*-1 | average reward: 22.9 -> 12.5 | Hurt |
| use NeuAgent (d) instead of humans | NeuAgent · WikiNav-16-4 | average reward: 5.0 -> 28.2 | Helped |
| pretrain on WikiNav-16-4 before fine-tuning | NeuAgent-Rec · WikiNav-Jeopardy | Recall@1: 13.9 -> 18.9 | Helped |
| pretrain on WikiNav-16-4 before fine-tuning | NeuAgent-Rec · WikiNav-Jeopardy | Recall@40: 33.2 -> 38.3 | Helped |
| pretrained NeuAgent instead of Lucene | NeuAgent-Rec · WikiNav-Jeopardy | Recall@4: 14.7 -> 23.6 | Helped |
Closest papers by meaning
- WebVLN: Vision-and-Language Navigation on Websites (AAAI 2024 · similarity 0.89)
- Empowering Efficiency and Efficacy in WebAgent via Enabling Info-Rich Seeking (ICLR 2026 · similarity 0.87)
- WebLINX: Real-World Website Navigation with Multi-Turn Dialogue (ICML 2024 · similarity 0.86)
- Go-Browse: Training Web Agents with Structured Exploration (ICLR 2026 · similarity 0.86)
- Language Agent Tree Search Unifies Reasoning, Acting, and Planning in Language Models (ICML 2024 · similarity 0.85)
- AutoGuide: Automated Generation and Selection of Context-Aware Guidelines for Large Language Model Agents (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.