MusicBrainz

该数据集是基于MusicBrainz数据库中的歌曲记录，并使用数据生成器创建具有修改属性值的副本作为基准。通过提供的URL，可以获取所有不同大小的变体。根据60/20/20的比例实施了训练/验证/测试的划分。数据规模为2万条记录（所使用的变体），任务为实体匹配。This dataset is built upon song records sourced from the MusicBrainz database, with a data generator utilized to create copies with modified attribute values as the benchmark dataset. All variants with different sizes are available via the provided URL. The train-validation-test split is conducted according to the 60/20/20 ratio. The dataset consists of 20,000 records (the employed variants), and the core task is entity matching.