Temporal knowledge graph reasoning using global and recent history information.

Area of Science:

• Artificial Intelligence
• Data Science
• Knowledge Representation

Background:

• Static Knowledge Graphs fail to capture real-world dynamic knowledge.
• Temporal Knowledge Graphs (TKGs) address temporal knowledge but face incompleteness challenges.
• Knowledge completion is crucial for reasoning and inferring missing facts in TKGs.

Purpose of the Study:

• To propose a novel event forecasting model, the Global-Recent Historical Network (GRHNet).
• To enhance knowledge completion in Temporal Knowledge Graphs by capturing historical event dynamics.
• To improve the prediction of future events by leveraging both global and recent historical information.

Main Methods:

• Developed GRHNet, a model simulating event evolution using statistical methods.
• Designed a global history learner to identify repeatable knowledge patterns.
• Implemented a recent history learner to capture time-varying knowledge dynamics.

Main Results:

• GRHNet was evaluated on two benchmark datasets.
• The model demonstrated superior performance compared to state-of-the-art baseline methods.
• Achieved at least a 3% relative improvement in Mean Reciprocal Rank (MRR).

Conclusions:

• GRHNet effectively captures both global repeatability and recent variability of knowledge.
• The proposed model offers a significant advancement in Temporal Knowledge Graph completion and event forecasting.
• GRHNet provides a robust approach for predicting future events with enhanced accuracy.