Towards Creating a New Triple Store for Literature-Based Discovery

Literature-based discovery (LBD) is a field of research aiming at discovering new knowledge by mining scientific literature. Knowledge bases are commonly used by LBD systems. SemMedDB, created with the use of SemRep information extraction system, is the most frequently used database in LBD. However, new applications of LBD are emerging that go beyond the scope of SemMedDB. In this work, we propose some new discovery patterns that lie in the domain of Natural Products and that are not covered by the existing databases and tools. Our goal thus is to create a new, extended knowledge base, addressing limitations of SemMedDB. Our proposed contribution is three-fold: 1) we add types of entities and relations that are of interest for LBD but are not covered by SemMedDB; 2) we plan to leverage full texts of scientific publications, instead of titles and abstracts only; 3) we envisage using the RDF model for our database, in accordance with Semantic Web standards. To create a new database, we plan to build a distantly supervised entity and relation extraction system, employing a neural networks/deep learning architecture. We describe the methods and tools we plan to employ.