Mind2Web

Mind2Web专注于为Web通用智能体的开发和评估提供支持，它包含超过2000个来自137个网站的开放式任务，这些网站覆盖了31个领域，并为这些任务提供了众包的操作序列。数据集中的每个任务都包含唯一的ID、网站名称、领域、子领域、任务描述、操作序列的可读字符串表示以及操作列表。每个操作步骤包括操作类型（如点击、输入、选择）和相关的值。数据集采用CC-BY 4.0许可协议，并且声明仅用于研究目的，以期通过语言技术提高Web的可访问性。Mind2Web is dedicated to supporting the development and evaluation of general-purpose web AI Agents. It contains over 2,000 open-ended tasks sourced from 137 websites spanning 31 distinct domains, with crowdsourced action sequences provided for each task. Each task in the dataset includes a unique ID, website name, domain, sub-domain, task description, human-readable string representation of the action sequence, and the action list. Each action step comprises an action type (e.g., click, input, selection) and relevant associated values. The dataset is released under the CC-BY 4.0 license, and it is explicitly stated that it is intended solely for research purposes, with the goal of improving web accessibility through language technologies.