Climate Data Literature Ontology and Entity Recognition System

The team built a comprehensive system to organize climate research literature and identify datasets referenced in scientific text. The solution includes a climate data literature ontology that captures relationships between institutions, sources, experiments, activities, citations, and licenses. The system also uses entity recognition to identify climate datasets mentioned in scientific papers, helping researchers understand which data sources are used in climate publications. A third component extracts and structures publication metadata from websites.

The team parsed input files to extract climate data attributes and map relationships. They built an entity recognition model using SciBERT, a language model specialized for scientific text, and trained it across multiple hyperparameters. They evaluated model performance using the SeqEval metric. For metadata extraction, they analyzed website structure and implemented parsing to capture publication information in a structured format.

The solution helps climate researchers navigate the landscape of climate data and literature. By structuring relationships in an ontology and identifying datasets in publications, the system enables more efficient retrieval and analysis of climate research materials.