OPENPI2.0

OPENPI2.0是由宾夕法尼亚大学开发的数据集，专注于文本中实体状态变化的追踪。该数据集包含55个来自wikiHow的日常活动程序，每个程序平均有5.0步。数据集通过众包方式标注了实体、属性和状态变化，并特别改进了实体和属性的规范化以及实体显著性标注，以提高模型评估的准确性。OPENPI2.0的应用领域包括问答和经典规划，旨在通过追踪显著实体的状态变化来提升下游任务的性能。

OPENPI2.0 is a dataset developed by the University of Pennsylvania, focusing on tracking state changes of entities in text. This dataset contains 55 daily activity procedures sourced from wikiHow, with an average of 5.0 steps per procedure. The dataset employs crowdsourcing to annotate entities, attributes and state changes, and has specifically optimized the normalization of entities and attributes as well as entity saliency annotation to improve the accuracy of model evaluation. The application scenarios of OPENPI2.0 include question answering and classical planning, aiming to enhance the performance of downstream tasks by tracking the state changes of salient entities.