Output details
11 - Computer Science and Informatics
University of Edinburgh
Datasets for generic relation extraction
<22> Originality: This paper rationalises some of the concepts in the task of relation extraction (a sub-task of information extraction). A publicly available corpus (the ACE corpus) is converted into a form for training and testing of relation extraction systems. Its utility is demonstrated through a series of experiments that compare approaches to generic relation extraction.
Significance: Clarifies many issues in relation extraction and provides a revised ACE dataset to other researchers (distributed by the Linguistic Data Consortium, the main data provider to NLP researchers across the world).
Rigour: Formal evaluation of a relation extraction system on the new dataset.