Using Transformer Networks and Knowledge Graphs in Earth Science Literature to Synthesize Mass Information for Transdisciplinary Research

Abstract:

Advancements in technology have ushered in a productive wave of transdisciplinary research in Earth Science and Machine Learning (ML). Both domains of research have vast amounts of knowledge within their respective bodies of scientific literature; synthesizing this information can lead to effective and efficient transdisciplinary research. We utilize pre-trained Transformer models, namely SpERT by Eberts and Ulges, to perform named entity recognition (NER) and relation extraction (RE) in scientific literature. NER involves the identification and classification of entities within text, while RE involves identifying and describing relationships between those entities. Extracted entities are categorized based on the schema laid out in the SciERC dataset by Luan et al., which defines six categories based on scientific literature: Method, Material, Task, Metric, General, or Other Scientific Term. Using the evaluation output of the SpERT model on scientific literature, we populate a knowledge graph data structure. We then visualize this network so the end user, transdisciplinary researchers, may synthesize mass information in scientific literature across multiple domains.