Who Should I Engage With at What Time? A Missing Event-Aware Temporal Graph Neural Network

Area of Science:

• Artificial Intelligence
• Machine Learning
• Network Science

Background:

• Temporal graph neural networks (GNNs) are increasingly vital for analyzing dynamic network data in fields like bioinformatics and social networks.
• Existing GNNs primarily focus on future event prediction, assuming complete historical data, which is often unrealistic.
• The accurate estimation of event timing is as crucial as predicting future events in many real-world applications.

Purpose of the Study:

• To develop a novel temporal GNN capable of handling unobserved events and accurately predicting both future occurrences and their timing.
• To uniformly model the evolving graph structure and event timing, including missing events, for enhanced predictive power.

Main Methods:

• Propose MTGN (Missing Temporal Graph Network), a GNN framework designed to be aware of missing events.
• Model the dynamics of both observed and missing events using two coupled temporal point processes (TPPs).
• Integrate the influence of missing events into the network's learning process.

Main Results:

• MTGN significantly outperforms existing temporal GNN methods on real-world temporal graph datasets.
• Achieved up to 89% improvement in time prediction accuracy.
• Demonstrated up to 112% improvement in link prediction accuracy.

Conclusions:

• MTGN offers a robust solution for temporal graph analysis by explicitly accounting for missing events.
• The proposed method enhances the prediction of future events and their precise timing in dynamic networks.
• MTGN provides a significant advancement in handling the complexities of real-world temporal graph data.