PUB - Publikationen an der Universität Bielefeld

Relation extraction is frequently and successfully addressed by machine learning methods. The downside of this approach is the need for annotated training data, typically generated in tedious manual, cost intensive work. Distantly supervised approaches make use of weakly annotated data, which can be derived automatically. Recent work in the biomedical domain has applied distant supervision for protein-protein interaction (PPI) with reasonable results, by employing the IntAct database. Training from distantly labeled corpora is more challenging than from manually curated ones, as such data is inherently noisy. With this paper, we make two corpora publicly available to the community to allow for comparison of different methods that deal with the noise in a uniform setting. The first corpus is addressing protein-protein interaction (PPI), based on named entity recognition and the use of IntAct and KUPS databases, the second is concerned with drug-drug interaction (DDI), making use of the database DrugBank. Both corpora are in addition labeled with 5 state-of-the-art classifiers trained on annotated data, to allow for development of filter methods. Furthermore, we present in short our approach and results for distant supervision on these corpora as a strong baseline for future research.