TransNet: Transfer Knowledge for Few-shot Knowledge Graph Completion

TL;DR

TransNet addresses few-shot knowledge graph completion under long-tail relation distributions by transferring knowledge across related tasks. It combines a relational message-passing GNN-based task similarity measure with a transformer-based Meta Relation Learner to produce a meta-relational representation, and a SkipTransD scoring module for relation-conditioned link prediction, all guided by task-conditioned meta-learning and contrastive relation refinement. The approach includes a WL-edge subtree kernel for task similarity, a dynamic adaptation factor $\alpha_r=\sigma(W_r^\top \psi(S_r))$, and a warm-up phase to mitigate negative transfer, with evaluation on NELL-One and Wiki-One showing consistent gains in MRR and Hits$@k$ over state-of-the-art baselines. These contributions enable robust cross-task transfer in KG completion and offer practical improvements for incomplete graphs.

Abstract

Knowledge graphs (KGs) are ubiquitous and widely used in various applications. However, most real-world knowledge graphs are incomplete, which significantly degrades their performance on downstream tasks. Additionally, the relationships in real-world knowledge graphs often follow a long-tail distribution, meaning that most relations are represented by only a few training triplets. To address these challenges, few-shot learning has been introduced. Few-shot KG completion aims to make accurate predictions for triplets involving novel relations when only a limited number of training triplets are available. Although many methods have been proposed, they typically learn each relation individually, overlooking the correlations between different tasks and the relevant information in previously trained tasks. In this paper, we propose a transfer learning-based few-shot KG completion method (TransNet). By learning the relationships between different tasks, TransNet effectively transfers knowledge from similar tasks to improve the current task's performance. Furthermore, by employing meta-learning, TransNet can generalize effectively to new, unseen relations. Extensive experiments on benchmark datasets demonstrate the superiority of TransNet over state-of-the-art methods. Code can be found at https://github.com/lihuiliullh/TransNet/tree/main

Figures (2)

• Figure 1: Architecture of our proposed TransNet Model.
• Figure 2: The shared embedding space contains pre-trained embeddings of entities and relations by the training data of KGC. Best viewed in color.

Theorems & Definitions (4)

• Definition 1
• Definition 2
• Definition 3
• Definition 4