WPRM/minibench-mm-mind2web

这是一个包含网页操作和标注信息的 dataset，其中包括原始HTML、清理后的HTML、操作类型、正候选和负候选序列、网站信息、域名、子域名、标注ID、确认的任务类型、截图、动作表示、目标动作索引、目标动作表示、从清理后的HTML生成的axtree字符串、只包含选择竞价的axtree字符串、候选竞价中的axtree字符串、可见竞价、包含过滤后的可见竞价的序列、在裁剪中的可见axtree、问题、目标、目标动作的HTML、目标动作竞价字符串、原始Mind2Web指令、裁剪坐标、可见元素数量、目标元素在裁剪中的可见性、原始截图、裁剪后的截图、带有som覆盖的所有图像、带有som覆盖的裁剪图像、在候选竞价中带有竞价的som覆盖、在带有竞价的候选竞价中裁剪的som覆盖、带有竞价和选择的som覆盖、目标元素是否包含在候选竞价中、目标元素是否包含在带选择的竞价中。数据集分为训练集和多个测试集，每个测试集专注于不同的任务或网站。

This dataset includes web action and annotation information, featuring original HTML, cleaned HTML, operation types, positive and negative candidate sequences, website information, domain, subdomain, annotation ID, confirmed task type, screenshots, action representations, target action index, target action representations, axtree strings from cleaned HTML, axtree strings with choice bidding only, candidate bidding axtree strings, visible bids, filtered visible bid sequences, visible axtree in crop, questions, targets, target action HTML, target action bidding strings, original Mind2Web instructions, crop coordinates, visible elements count, target element visibility in crop, original screenshots, cropped screenshots, som overlay images for all, som overlay images for crops, som overlay with bid in candidate bids, som overlay with bid in candidate bids cropped, som overlay with bid and choices, whether the target element is included in candidate bids, whether the target element is included in choice bids. The dataset is split into a training set and multiple test sets, each focusing on different tasks or websites.