umbc-scify/scifact-open

该数据集包含两个部分：distractors和test。distractors部分包含约500k个干扰项证据及其ID，用于训练。test部分包含206个主张及其相应的证据、证据ID、标签以及所有可能的证据和标签列表，用于测试。数据集旨在评估事实检索性能。

The dataset consists of two parts: distractors and test. The distractors part includes approximately 500k distractor evidence items and their IDs for training. The test part includes 206 claims with their corresponding evidence, evidence IDs, labels, and lists of all possible evidence and labels for testing. The dataset is designed to evaluate fact retrieval performance.