A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis
This addresses the challenge of autonomous web automation for users needing to interact with complex websites, representing a strong specific gain rather than a broad paradigm shift.
The paper tackles the problem of poor performance of LLM-driven agents on real-world websites due to open domainness, limited context length, and lack of inductive bias on HTML, by introducing WebAgent, which improves success on real websites by over 50% and achieves an 18.7% higher success rate than prior methods on the MiniWoB benchmark.
Pre-trained large language models (LLMs) have recently achieved better generalization and sample efficiency in autonomous web automation. However, the performance on real-world websites has still suffered from (1) open domainness, (2) limited context length, and (3) lack of inductive bias on HTML. We introduce WebAgent, an LLM-driven agent that learns from self-experience to complete tasks on real websites following natural language instructions. WebAgent plans ahead by decomposing instructions into canonical sub-instructions, summarizes long HTML documents into task-relevant snippets, and acts on websites via Python programs generated from those. We design WebAgent with Flan-U-PaLM, for grounded code generation, and HTML-T5, new pre-trained LLMs for long HTML documents using local and global attention mechanisms and a mixture of long-span denoising objectives, for planning and summarization. We empirically demonstrate that our modular recipe improves the success on real websites by over 50%, and that HTML-T5 is the best model to solve various HTML understanding tasks; achieving 18.7% higher success rate than the prior method on MiniWoB web automation benchmark, and SoTA performance on Mind2Web, an offline task planning evaluation.