%0 Generic %1 gur2023realworld %A Gur, Izzeddin %A Furuta, Hiroki %A Huang, Austin %A Safdari, Mustafa %A Matsuo, Yutaka %A Eck, Douglas %A Faust, Aleksandra %D 2023 %K html llm ml transformer webagent %T A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis %U http://arxiv.org/abs/2307.12856 %X Pre-trained large language models (LLMs) have recently achieved better generalization and sample efficiency in autonomous web navigation. 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 can complete the 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 generated Python programs 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 recipe improves the success on a real website by over 50%, and that HTML-T5 is the best model to solve HTML-based tasks; achieving 14.9% higher success rate than prior SoTA on the MiniWoB web navigation benchmark and better accuracy on offline task planning evaluation.

@misc{gur2023realworld, abstract = {Pre-trained large language models (LLMs) have recently achieved better generalization and sample efficiency in autonomous web navigation. 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 can complete the 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 generated Python programs 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 recipe improves the success on a real website by over 50%, and that HTML-T5 is the best model to solve HTML-based tasks; achieving 14.9% higher success rate than prior SoTA on the MiniWoB web navigation benchmark and better accuracy on offline task planning evaluation.}, added-at = {2023-08-07T18:19:20.000+0200}, author = {Gur, Izzeddin and Furuta, Hiroki and Huang, Austin and Safdari, Mustafa and Matsuo, Yutaka and Eck, Douglas and Faust, Aleksandra}, biburl = {https://www.bibsonomy.org/bibtex/2007105d2179df8e22bb507d11a767a86/talhankoc}, description = {A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis}, interhash = {b7d3ac1d5c43efc31de3dfe8b1adc1b0}, intrahash = {007105d2179df8e22bb507d11a767a86}, keywords = {html llm ml transformer webagent}, note = {cite arxiv:2307.12856}, timestamp = {2023-08-07T18:19:20.000+0200}, title = {A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis}, url = {http://arxiv.org/abs/2307.12856}, year = 2023 }