Extracting bacteria biotopes with semi-supervised named entity recognition and coreference resolution

  • Authors:
  • Nhung T. H. Nguyen;Yoshimasa Tsuruoka

  • Affiliations:
  • Japan Advanced Institute of Science and Technology, Asahidai, Nomi, Ishikawa, Japan;Japan Advanced Institute of Science and Technology, Asahidai, Nomi, Ishikawa, Japan

  • Venue:
  • BioNLP Shared Task '11 Proceedings of the BioNLP Shared Task 2011 Workshop
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper describes our event extraction system that participated in the bacteria biotopes task in BioNLP Shared Task 2011. The system performs semi-supervised named entity recognition by leveraging additional information derived from external resources including a large amount of raw text. We also perform coreference resolution to deal with events having a large textual scope, which may span over several sentences (or even paragraphs). To create the training data for coreference resolution, we have manually annotated the corpus with coreference links. The overall F-score of event extraction was 33.2 at the official evaluation of the shared task, but it has been improved to 33.8 thanks to the refinement made after the submission deadline.