Changelog

4.1.1

Applied the changes from 4.0.1 to 4.1.0.

Data within the court and government_gazzette subsets was anonymised using models trained from the MAPA Project. While the data in this subset was obtained from the public domain, it was still anonymised to safeguard the identity of individuals who are not Politically Exposed Persons. The following entities were processed: surnames, ID document numbers, and cities (for addresses) Anonymisation was performed at the document level to retain its coherence, so if an entity is replaced, it is replaced the same way within a document.

The shuffled data was regenerated with any changed data. The train/validation/test splits are not kept the same as the 4.0.* version.

4.0.1

Fix dataset metadata, due to change in the datasets library. As part of this, the dataset metadata is now directly available when loading the dataset, not just the raw files as in the previous version.

4.0.0

Initial release of the corpus, used to train BERTu.