DebateLabKIT/argument_mapping_dpo_pairs

该数据集包含了BBC新闻、企业游说、日常困境、药物评价和美国最高法院案例等五个不同的主题。每个主题下都有选中的(chosen)和拒绝的(rejected)内容及其角色(role)。数据集以训练集的形式提供，并且每个主题都有对应的文件路径。

The dataset consists of five configurations: bbc_news, corporate_lobbying, daily_dilemmas, drug_reviews, and scotus. Each configuration includes two features: chosen and rejected, with each feature having two fields: content and role, both of which are strings. The dataset also includes information on the splits for the training set, as well as the size of the dataset and the download size. The dataset is provided in the form of training sets, with each topic having its corresponding file path.