The dataset is crawled from EUR-Lex, which is a corpus of legislative documents of the European Union. It contains many different types of documents like treaties, legislation, case-law, legislative proposals etc. which are in twenty-four official European languages. The dataset constitutes a very challenging multilabel scenario due to the high number of around 4000 labels and 20000 documents.
The main challenge in multi label classification is that regular classification methods cannot be employed since several labels could be possibly assigned to a single document. Incorporating a model that could perform well in this scenario of large-scale legal domain was the major motive.
Exploratory Analysis of EurLex Dataset-
