Extracting Drug-Drug Interaction from Literature through Detecting Linguistic-Based Negation and Neutral Candidates

Extracting biomedical relations such as drug-drug interaction (DDI) from text is an important task in biomedical NLP. Most of the current relation extraction tasks and the produced corpora are based on binary relations, which decide whether a relation between the two entities exists in the sentence or not. The paper aims to identify neutralinteraction candidates who have not been stated by the authors in the sentences which have not been studied yet. Alongside, a linguistic-based negation related feature set based on negation scope and cues were employed. The proposed neutral candidate features show significantly better performance in comparison with negation related features. The results show by employing the proposed features alongside to a bag of words kernel, the performance of the three used kernel methods improves. Moreover, the enhanced local context kernel outperformed the other used methods. Additionally the experiments show the automatically produced negation scope and cue tags can be employed effectively.

Keywords:

Drug-Drug interaction , kernel methods , negation detection , neutral candidate