Leveraging document-specific information for classifying relations in scientific articles

Abstract

Tremendous amount of knowledge is present in the ever-growing scientific literature. In order to grasp this massive amount knowledge, various computational tasks are proposed for training computers to read and analyze scientific documents. As one of these task, semantic relationship classification aims at automatically analyzing semantic relationships in scientific documents. Conventionally, only a limited number of commonly used knowledge bases such as Wikipedia are used for collecting background information for this task. In this work, we hypothesize that scientific papers also could be utilized as a source of background information for semantic relationship classification. Based on the hypothesis, we propose the model that is capable of extracting background information from unannotated scientific papers. Preliminary experiments on the RANIS dataset [1] proves the effectiveness of the proposed model on relationship classification in scientific articles.