Analyzing Deep Learning Methods for Constructing Knowledge Graphs From Text Using Automated Ontology

Abstract

There already exist many different methods for constructing knowledge graphs from text-based inputs. While it has been concluded that deep learning approaches are best for natural language processing-based ontology, the field, specifically in academia, is still developing. Specifically, it seems that with academic language such as biomedicine, NLP may need specialization because of the unique words and phrases used.

Our team explored different facets involving the generation of knowledge graphs using informational text with a dataset of over 7 million sentences from Wikipedia. My work distinctively involved observing the effect that the volume of information had on the overall accuracy of the generated ontology, the measuring of which involved the manual labeling of a subset of results.