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.
展开▼