Context-Aware Learning and Pattern Decomposition for Temporal Knowledge Graph Reasoning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Graph neural networks (GNNs) excel in temporal knowledge graph (TKG) reasoning.
• Existing methods struggle with local contextual dynamics and emerging patterns in evolving TKGs.

Purpose of the Study:

• To address limitations in modeling local dynamics and handling novel interactions in TKGs.
• To propose TCDR-PD, a network for enhanced temporal and contextual dynamic representation with pattern decomposition.

Main Methods:

• Introduced a temporal and contextual dynamic representation learning (TCDR) module for global trends and query-specific dynamics.
• Developed a pattern decomposition (PD) prediction module to disentangle recurring and emerging patterns.
• Evaluated TCDR-PD on four benchmark datasets for TKG reasoning.

Main Results:

• TCDR-PD demonstrated superior performance compared to state-of-the-art methods.
• The TCDR module effectively captured both temporal trends and contextual dynamics.
• The PD module successfully handled recurring and emerging patterns, improving prediction accuracy.

Conclusions:

• TCDR-PD offers a robust solution for stable reasoning over evolving temporal knowledge graphs.
• The proposed approach enhances adaptability to dynamic environments and novel interactions.
• This work advances the field of TKG reasoning by addressing key challenges in representation learning and prediction.