Few-shot temporal knowledge graph completion based on meta-optimization

Abstract

Abstract Knowledge Graphs (KGs) have become an increasingly important part of artificial intelligence, and KGs have been widely used in artificial intelligence fields such as intelligent answering questions and personalized recommendation. Previous knowledge graph completion methods require a large number of samples for each relation. But in fact, in KGs, many relationships are long-tail relationships, and the existing researches on few-shot completion mainly focus on static knowledge graphs. In this paper, we consider few-shot completion in Temporal Knowledge Graphs (TKGs) where the event may only hold for a specific timestamp, and propose a model abbreviated as FTMO based on meta-optimization. In this model, we combine the time-based relational-aware heterogeneous neighbor encoder, the cyclic automatic aggregation network, and the matching network to complete the few-shot temporal knowledge graph. We compare our model with the baseline models, and the experimental results demostrate the performance advantages of our model.

Keywords

• Long-tail relationships
• Meta-optimization
• Few-shot temporal knowledge graph