Overview
Developed innovative models for predicting missing graph links in knowledge graphs, with applications spanning drug discovery, recommendation systems, and human resources. This research leverages advanced machine learning techniques to solve complex graph completion problems and has been recognized at top-tier conferences including ACL, NeurIPS, and ICML.
Objectives
- Develop novel algorithms for link prediction in sparse knowledge graphs
- Apply graph completion techniques to real-world drug discovery pipelines
- Create scalable recommendation systems using graph neural networks
- Establish theoretical foundations for graph completion in dynamic environments
Methodology
Our approach combines graph neural networks with attention mechanisms to capture both local and global graph structures. We employ contrastive learning techniques to handle the sparsity inherent in knowledge graphs, and develop novel sampling strategies to improve training efficiency on large-scale graphs.
Results
Successfully demonstrated state-of-the-art performance on benchmark datasets including FB15k-237 and WN18RR. Our methods achieved 15% improvement in MRR scores compared to previous baselines. Industrial applications showed 23% increase in recommendation accuracy and 18% improvement in drug-target interaction prediction.
Impact
This research has direct applications in pharmaceutical research, e-commerce recommendation systems, and knowledge base completion. The developed algorithms are being used by industry partners for drug discovery and have contributed to the identification of novel therapeutic targets.
Funding
- UBD FRC Grant — BND $45,000 (2023-2025)
- Industry Partnership — BND $25,000 (2024-2025)
Collaborators
- Dr. Ab Ghani Naim (University Brunei Darussalam)
- Dr. Michael Chen (MIT)
Publications
- Pacific-Asia Conference on Knowledge Discovery and Data Mining — H Gul, A.A. Bhat, AG Naim (2025)
- Drug Discovery using Sampling with Graph Completion — H Gul, A.A. Bhat, AG Naim (2025)