WPRM/minibench-multimodal-mind2web

该数据集包含了用户在网站上的行为信息，以及与网页操作相关的HTML内容。具体包括行为唯一标识符、原始HTML和清理后的HTML、操作类型、正负候选元素序列、网站及子域名信息、标注ID、任务确认信息、屏幕截图、行为表示、目标行为索引、目标行为表示、从清理后的HTML生成的AXTree字符串、仅含选择_bid的AXTree字符串、候选_bid的AXTree字符串、可见bid、过滤后包含在可见bid中的元素、在裁剪中的可见AXTree、问题、目标、目标行为HTML、目标行为bid字符串、原始mind2web指令、裁剪坐标、可见元素计数、目标元素在裁剪中的可见性、裁剪的屏幕截图、 SOM叠加图像、带有候选bid中bid的SOM叠加图像、目标元素是否包含在候选bid中、目标元素是否包含在选择_bid中。数据集分为三个测试集，分别是针对网站、域名和任务的测试集。

The dataset contains user behavior information on websites and related HTML content for web operations. It includes action unique identifier, raw HTML and cleaned HTML, operation type, positive and negative candidate element sequences, website and subdomain information, annotation ID, confirmed task, screenshot, action representation, target action index, target action representation, AXTree string from cleaned HTML, AXTree string with only choice_bid, AXTree candidate_bid string, visible bids, filtered visible bid included, AXTree visible in crop, question, target, target action HTML, target action bid string, original mind2web instruction, crop coordinates, visible elements count, target element visible in crop, cropped screenshot, SOM overlay image, SOM overlay image with bid in candidate bids, target element included in candidate bids, target element included in choices_bid. The dataset is split into three test sets focusing on websites, domains, and tasks respectively.