Knowledge Graph Completion

Knowledge graphs (KGs) frequently exhibit incompleteness and sparsity, which significantly constrain their effectiveness for downstream applications. Although existing methods attempt to mitigate these limitations through either structural embedding techniques or pre-trained language models (PLMs), they face persistent challenges with few-shot relations and demonstrate limited success in effectively combining structural and semantic knowledge. To overcome these limitations, we propose two novel frameworks: (1) a relation-aware meta-learning approach for few-shot KG completion, and (2) a structure-semantic fusion method bridging PLMs and KG embeddings.

Committee: Dr. Qi Li (major professor), Dr. Hongyang Gao, Dr. Mengdi Huai, Dr. Wensheng Zhang and Dr. Ying Cai