Analyzing text in search of bio-molecular events

Abstract

The BioNLP'09 Shared Task on Event Extraction is a challenge which concerns the detection of bio-molecular events from text. In this paper, we present a detailed account of the challenges encountered during the construction of a machine learning framework for participation in this task. We have focused our work mainly around the filtering of false positives, creating a high-precision extraction method. We have tested techniques such as SVMs, feature selection and various filters for data pre- and post-processing, and report on the influence on performance for each of them. To detect negation and speculation in text, we describe a custom-made rule-based system which is simple in design, but effective in performance.