The School of Computer Science is pleased to present…
A Unified Framework for Continuous-Time Link Prediction in Temporal Knowledge Graphs
PhD Dissertation Proposal by: Amangel Bhullar
Date: 24 Feb. 2026
Time: 1:00pm-3:00pm
Location: OBB04 (Odette)
Abstract:
Knowledge graphs are widely used to represent structured relational data in applications such as recommendation systems, event forecasting, and decision support. A central task in these systems is link prediction — inferring missing or future relationships between entities. However, most existing temporal knowledge graph methods model time as discrete snapshots, even though real-world knowledge evolves continuously and irregularly. Furthermore, hierarchical structure, relational dependency, long-term temporal memory, and graph sparsity remain open challenges for reliable prediction. This dissertation proposes a unified framework for continuous-time link prediction in temporal knowledge graphs. The research progressively introduces five models addressing complementary limitations: TempHypE models smooth temporal evolution using hyperbolic neural ordinary differential equations; TempHypE-GNN incorporates relational neighborhood influence through geometry-aware message passing; SD-GNN enables scalable prediction via adaptive sparsity; RHGNN maintains stable long-term temporal memory; and KARMA introduces temporal anchors to infer missing and unobserved timestamps. Across standard temporal knowledge graph benchmarks, the proposed methods show consistent improvements in Mean Average Rank (MAR), Mean Reciprocal Rank (MRR), Hits@10, Hits@3, Hits@1 of prediction ranks over time and improved forecasting at unobserved timestamps.
Thesis Committee:
Internal Reader: Dr. Dan Wu
Internal Reader: Dr. Hamidreza Koohi
External Reader: Dr. Esam Abdel-Raheem
Advisor: Dr. Ziad Kobti
Registration Link (only MAC students need to pre-register)
