Master’s Thesis: PubMed MeSH Annotation with Graph Neural Networks

Project overview

Completed as my Master’s thesis at the University of Stuttgart and carried out at QUIBIQ GmbH, this work explored using graph neural networks in a link-prediction setting to assign MeSH headings to PubMed abstracts. The goal was to make use of graph-structured information that transformer-based models might overlook.

Approach and analysis

• Applied a GNN link-prediction approach to MeSH annotation.
• Analyzed prediction errors and a plausible reason for the model’s weaker performance.

Supervision and support

• Main supervisors: Prof. Dr. Roman Klinger and Prof. Dr. Steffen Staab
• Direct supervisors: Dr. Rafika Boutalbi and Anastasiia Iurshina
• Industry research guide: Dr. Juan G. Díaz Ochoa
• Funding: QUIBIQ GmbH, with Dr. Felix Weil

Publication

Annotating PubMed Abstracts with MeSH Headings using Graph Neural Network

Published in the Proceedings of the Fourth Workshop on Insights from Negative Results in NLP (2023). Code · Publications page