Metadata, or data about data, refers to information that provides structured context and details about other data. In the context of scientific publications, metadata can facilitate the discovery process of those publications and help researchers in exploring the enormous landscape of scientific literature.
In this thesis, we delve into the exploration of techniques for the extraction and parsing of essential metadata, such as article authors and references, from scientific publications. We propose our novel and effective approach to extract metadata both from the article itself and from the references within the publication.
Our tool, BiBEx, is designed to accurately extract metadata from both historical and contemporary scholarly literature, encompassing texts in both English and German. To accomplish this, we curated two distinct datasets: one dedicated to document segmentation and the other focused on reference segmentation and parsing.
Our results demonstrate that BiBEx achieves comparable or slightly improved performance when compared to commonly used tools for metadata extraction in scientific literature.
This project is divided in three parts
- Code contains all applications, Jupyter Notebooks, and Code fragments.
- Datasets contains the final datasets created in the context of this thesis.
- LaTeX contains all files associated with written parts of this thesis.
Note: This project does not include all files (binaries, pdfs, intermediate files, etc.) related to this thesis. Please refere to the readme file of a folder to download all files associated with this project.
Datasets and ML Models created and used in this thesis can be found on the GEOcite huggingface page.
A demonstrator of the BiBEx application can be found here.