Temporal Tagging of Noisy Clinical Texts in Brazilian Portuguese

Abstract

Temporal expressions are present in several types of texts, including clinical ones. The current research over temporal expressions has been done by the use of rule-based systems, machine learning or hybrid approaches, in most cases, over annotated (labeled) news texts correctly written in English. In this paper, we propose a method to extract and normalize temporal expressions from noisy and unlabeled clinical texts (discharge summaries) written in Brazilian Portuguese using a rule-based approach. The obtained results are similar to the state-of-the-art researches made with the same purpose in other languages. The proposed method reached a F1 score of 88.92% for the extraction step and, a F1 score of 87.89% for the normalization step.