HiEve_Dataset

This repo is a cleaned version of HiEve Dataset.
You can find 100 documents in hievents_v2/processed. Each document contains multiple lines, e.g., Text, Event, and Relation (each followed by "\t" and contents).

Illustration for line Event / Relation:
Event    1    competing     occurrence    98
Event    id    mention    type    char_ID
Relation    1    2    Coref    true    competing    race
Relation    id1    id2    relation    uncertain    mention1    mention2
(Note: the uncertain slot can be ignored)

Transitive closure has been applied to generate labels for relations not annotated in the dataset (as shown in HiEve_Preprocess.ipynb).
You might downsample NoRel by a certain ratio (e.g., 40%) following previous work.