New resources and perspectives for biomedical event extraction

  • Authors:
  • Sampo Pyysalo;Pontus Stenetorp;Tomoko Ohta;Jin-Dong Kim;Sophia Ananiadou

  • Affiliations:
  • National Centre for Text Mining and University of Manchester, Manchester Interdisciplinary Biocentre, Manchester, UK;Tokyo University, Hongo, Bunkyo-ku, Tokyo, Japan;National Centre for Text Mining and University of Manchester, Manchester Interdisciplinary Biocentre, Manchester, UK;Database Center for Life Science, Yayoi, Bunkyo-ku, Tokyo, Japan;National Centre for Text Mining and University of Manchester, Manchester, UK

  • Venue:
  • BioNLP '12 Proceedings of the 2012 Workshop on Biomedical Natural Language Processing
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Event extraction is a major focus of recent work in biomedical information extraction. Despite substantial advances, many challenges still remain for reliable automatic extraction of events from text. We introduce a new biomedical event extraction resource consisting of analyses automatically created by systems participating in the recent BioNLP Shared Task (ST) 2011. In providing for the first time the outputs of a broad set of state-of-the-art event extraction systems, this resource opens many new opportunities for studying aspects of event extraction, from the identification of common errors to the study of effective approaches to combining the strengths of systems. We demonstrate these opportunities through a multi-system analysis on three BioNLP ST 2011 main tasks, focusing on events that none of the systems can successfully extract. We further argue for new perspectives to the performance evaluation of domain event extraction systems, considering a document-level, "off-the-page" representation and evaluation to complement the mention-level evaluations pursued in most recent work.