Improving scientific relation classification with task specific supersense

Classifying the relationship between entitiesis an important natural language processing(NLP) task. Scientific Relation Classificationaims at automatically categorizing scientificsemantic relationships among entities in scientific documents. Conventionally, only taskunspecific supersense, such as supersense (orhyernym) from WordNet (e.g., ANIMAL is thesupersense of "dog"), is used as a feature for relation classification. In this work, we hypothesize that task specific supersense could also beutilized as an informative feature for relationclassification. Specifically, we define a newentity type based on the property of a giventask, and facilitate scientific relation classification with the task specific supersense. Ourexperiments on three different datasets provethe effectiveness of the task specific supersenseon relation classification in scientific articles.