Linked Hypernyms Dataset

v2016-04

LHD Extraction Framework

The source code and documentation is available on github.

Resources

Development corpus: tagged corpus (in GATE NLP framework) used for development of extraction grammars. The corpus contains roughly 1800 documents/first sentences of Wikipedia articles/ with the occurrence of the first hypernym annotated (600 for English, 600 for German and 600 for Dutch).
Extraction grammars (en,de,nl) : JAPE grammars for Hearst pattern discovery in first senteces of Wikipedia articles for Enlish, German and Dutch.

hSVM Framework

The source code and documentation is available on github.

The Linked Hypernym Dataset is dual licensed under the Creative Commons Attribution Only (CC BY 3.0) license and Creative Commons Attribution Only Share Alike (CC BY-SA 3.0) license.
This dataset is used in a named entity recognition system called Targeted Hypernym Discovery.
The datasets are processed from Wikipedia snapshots made in March 2016.