Crowdsourced Saliency Evaluation Dataset

该数据集由800名群众工作者提供注释，他们通过执行文本分类任务，对七种显著性方法在两个数据集上的性能进行了评估。该数据集包含实例级数据，旨在用于评估自然语言处理中的显著性方法。规模上，大约有4000个任务，缺失率高达97.5%。这项任务是基于显著性解释的文本分类。

This dataset was annotated by 800 crowd workers, who evaluated the performance of seven saliency methods on two datasets by performing text classification tasks. It comprises instance-level data and is intended for the evaluation of saliency methods in natural language processing (NLP). Spanning approximately 4,000 tasks, this dataset has a missing rate as high as 97.5%. The task involved is saliency explanation-based text classification.