Legal-Docs-Large-MLTC

1. Prepare datasets

The following script will:

• Download JRC-Aquis data.
• Download and prepare EuroVoc data and EuroVoc Analysis tool.
• Prepare JRC-Aquis datasets for monolingual and multi-lingual experiments.

bash preparedata.sh

OUTPUT FILES

• JRC Monolingual Dataset: datasets/jrc_en_basic.csv
• JRC Multilingul Dataset: datasets/jrc_3langs_basic.csv
• EURLEX57K Dataset: datasets/EurLex57K.csv
• Pickle of EuroVocAnalyzeTool instance: data/EuroVocAnalysisTool.pickle
  • This instance store EuroVoc data.
  • This object provides access to relations within EuroVoc hierarchy.

TODO

• Provide splits directory from bash script.

2. Iterative split

Default splits for the datasets are provided in Iterative_Split/JRC_Aquis and Iterative_Split/EurLex57K To make another split using iterative split approach, follow instructions in Iterative_Split/README.md

TODO

• perepare and preprocess dataset
  • JRC-Aquis dataset
  • EurLex57K
• Hierachical Reduction / extract MT/TT/DO information
• LM Finetuning & Classifier Finetuning
  • AWD-LSTM
  • Transformers
  • bash scripts
• logs&results