MusicUGC-NER

This dataset contains user-generated content (UGC) containing music entities (work of arts and performing artists). The datasources of this dataset are Reddit and YouTube. The datasets consist each of a k-fold split (provided in .IOB, following the IOB format. Furthermore, we include the metadata of items in the respective data.jsonl files. All datasets can be found in data/dataset. We provide the following datasets: reddit+shsyt: The full MusicUGC-NER dataset. It contains annotations by us (YouTube) and by researchers from Deezer (Reddit). Please see citation information below and in the corresponding repository. shsyt: The YouTube-only subset. shs100k2: The automatically matched data based on versions in SHS100K2. While this dataset is rather big, the data quality is much less than the two other datasets, since it was not curated by humans and we observed a rather low recall (please refer to our conference paper for details).