A temporal knowledge graph reasoning model based on recurrent encoding and contrastive learning.

Area of Science:

• Artificial Intelligence
• Data Science

Background:

• Temporal knowledge graphs (TKGs) are essential for modeling time-evolving data in intelligent systems.
• Predicting future events in TKGs (extrapolation) is challenging, as existing models often neglect the impact of past occurrences.

Purpose of the Study:

• To develop a novel temporal knowledge graph reasoning model, TRCL, that enhances extrapolation accuracy.
• To address the limitations of current models in capturing historical influences on future predictions.

Main Methods:

• TRCL integrates recurrent encoding to capture temporal dynamics and entity/relationship evolution.
• A global historical matrix accounts for recurring past events.
• Contrastive learning minimizes interference from historical data during future event prediction.

Main Results:

• TRCL demonstrated superior performance on four benchmark datasets, outperforming state-of-the-art models.
• The model achieved a 1.03% improvement in Mean Reciprocal Rank (MRR) on the ICEWS14 dataset compared to the TiRGN baseline.

Conclusions:

• TRCL significantly enhances the accuracy and robustness of temporal knowledge graph extrapolation.
• The model establishes a new benchmark for dynamic knowledge graph applications and predictive intelligence systems.