Semantic–Structural Representation Learning for Dynamic Link Prediction in Temporal Graphs
PhD Dissertation Defense by: Nahid Abdolrahmanpour Holagh
Date: September 18
Time: 2:30 p.m
Location: Teams Meeting
Abstract:
Dynamic Link Prediction (DLP) aims to predict future interactions in evolving networks and supports applications such as social network analysis, recommendation systems, communication forecasting, and citation prediction. Although Graph Neural Networks and Graph Transformers have improved the modeling of temporal and structural dependencies, many existing approaches remain structure-centric and do not fully use the rich semantic information associated with graph entities.
This dissertation investigates semantic–structural representation learning for DLP in temporal graphs. It explores how textual information associated with nodes can be integrated with evolving graph structure to improve the prediction of future links. Particular attention is given to incorporating semantic information into temporal attention mechanisms and modeling semantic and structural evolution as interacting processes rather than treating semantics as static node features.
Experimental evaluations on real-world temporal networks demonstrate that combining semantic and structural information improves both classification and ranking performance compared with structure-only and static-semantic approaches. In conclusion, this research establishes a unified semantic–structural perspective for dynamic link prediction and provides a foundation for future work in temporal graph learning and language-enhanced graph representation learning.
Keywords: [Insert 3-5 keywords]
Dynamic Link Prediction, Temporal Graphs, Graph Transformers, Semantic–Structural Learning.
Doctoral Committee:
Internal Reader: Dr. Dan Wu
Internal Reader: Dr. Jianguo Lu
External Reader: Dr. Narayan Kar
External Examiner: Dr. Bijan Raahemi
Advisor(s): Dr. Ziad Kobti
Chair: Dr. Dennis Jackson
