Learning from Scholarly Attributed Graphs: A Graph Representation Learning Approach

Abstract

My doctoral thesis at King Abdullah University of Science and Technology presents a unified framework for learning from scholarly attributed graphs — citation networks where nodes carry content and structure simultaneously. The work spans three research threads: dataset mining and retrieval for scholarly search (Delve), semi-supervised and multi-label node classification via reinforcement-learning graph walks (RAW, CGW), and temporal graph learning for biomedical hypothesis generation (T-PAIR).