Datasets:

Modalities:
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:
osalaun commited on
Commit
3f48876
1 Parent(s): 6fd85a5

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +12 -4
README.md CHANGED
@@ -90,10 +90,18 @@ EUROPA is a dataset designed for training and evaluating multilingual keyphrase
90
 
91
  ## Dataset Structure
92
 
93
- - **celex_id:** CELEX identifier inherited from CJEU. Different translated versions of the same judgment share the same celex_id.
94
- - **lang:** ISO 639-1 language code
95
- - **input:** judgment transcription or translation
96
- - **keyphrases:** reference keyphrases drafted by the CJEU. Phrases are separated by semicolons.
 
 
 
 
 
 
 
 
97
 
98
  ## Citation
99
 
 
90
 
91
  ## Dataset Structure
92
 
93
+ - **celex_id:** CELEX identifier inherited from CJEU. Different translated versions of the same judgment share the same celex_id. If you wish to have a unique identifier for each instance, you can concatenate `lang` and `celex_id` values;
94
+ - **lang:** ISO 639-1 language code;
95
+ - **input:** judgment transcription or translation;
96
+ - **keyphrases:** reference keyphrases drafted by the CJEU.
97
+
98
+ This page presents an randomly split version of the dataset, thus allowing all sets to have the same distribution in terms of vocabulary and languages. More details can be found in Appendix H of our paper.
99
+
100
+ - **training set**: 159 306 instances;
101
+ - **validation set**: 53 943 instances;
102
+ - **test set**: 71 708 instances.
103
+
104
+
105
 
106
  ## Citation
107