Few-Shot Knowledge Graph Completion via Transfer Knowledge from Similar Tasks

Abstract

Knowledge graphs (KGs) are widely used in many AI applications, but they are often incomplete, which limits their effectiveness. In many cases, most relations have very few examples, making it difficult to learn accurate models. Few-shot learning has emerged as a promising solution by enabling KG completion with only a small number of training triplets. However, many existing methods treat each relation separately and miss the opportunity to share useful information across tasks. In this paper, we propose TransNet, a transfer learning approach for few-shot KG completion that captures relationships between tasks and reuses knowledge from related ones. TransNet also uses meta learning to better handle unseen relations. Experiments on standard benchmarks show that TransNet achieves strong performance compared to prior methods. Code and data will be released upon acceptance.