Multi-hop interpretable meta learning for few-shot temporal knowledge graph completion.

Area of Science:

• Artificial Intelligence
• Data Science

Background:

• Traditional multi-hop path completion models struggle with temporal knowledge graphs, especially for few-shot relations.
• Existing models fail to account for timestamp information's impact on path completion.

Purpose of the Study:

• To propose a Few-shot Temporal knowledge graph completion model based on Multi-hop Interpretable meta-learning (FTMI).
• To enhance multi-hop path completion for temporal knowledge graphs with limited relationship instances.

Main Methods:

• FTMI aggregates multi-hop neighbor information for time-aware entity representations.
• Transformer encoding and mean pooling are used for entity-pair and relationship representations.
• Reinforcement learning with a novel reward function optimizes path selection for completion.

Main Results:

• The FTMI model effectively incorporates timestamp information to understand temporal dynamics.
• Meta-learning enables rapid adaptation to new tasks with few samples.
• Experimental validation on two datasets demonstrates the model's effectiveness.

Conclusions:

• FTMI significantly improves few-shot temporal knowledge graph completion.
• The model offers a robust solution for handling sparse temporal relationship data.
• FTMI provides interpretable completion results by considering temporal context.