Drug-Drug Interaction Extraction from Biomedical Texts with SVM and RLS Classifiers

First Challenge Task on Drug-Drug Interaction Extraction 2011.

医药卫生;计算机科学

Jari Bjo¨rne ; 1 ; 2 Antti Airola ; 1 ; 2 Tapio Pahikkala1 ; Tapio Salakoski1

Others  :  http://ceur-ws.org/Vol-761/paper4.pdf
PID  :  42353

来源: CEUR

【 摘 要 】

We introduce a system developed to extract drug-drug in-teractions (DDI) for drug mention pairs found in biomedical texts. This system was developed for the DDI Extraction First Challenge Task 2011 and is based on our publicly available Turku Event Extraction System, which we adapt for the domain of drug-drug interactions. This system relies heavily on deep syntactic parsing to build a representation of the relations between drug mentions. In developing the DDI extraction sys- tem, we evaluate the suitability of both text-based and database derived features for DDI detection. For machine learning, we test both support vector machine (SVM) and regularized least-squares (RLS) classifiers, with detailed experiments for determining the optimal parameters and training approach. Our system achieves a performance of 62.99% F-score