A Novel Temporal Network-Embedding Algorithm for Link Prediction in Dynamic Networks.

Area of Science:

• Network Science
• Machine Learning
• Graph Representation Learning

Background:

• Understanding complex systems like human interactions and biological networks is crucial.
• Predicting future links in dynamic networks has significant practical applications.
• Existing network embedding models often neglect the temporal dynamics of networks.

Purpose of the Study:

• To enhance the understanding of network evolution by addressing the link-prediction problem in temporal networks.
• To develop an advanced machine learning approach using graph representation learning for temporal networks.
• To propose a novel temporal network-embedding algorithm that captures the evolving nature of networks.

Main Methods:

• Developed a novel temporal network-embedding algorithm, TempNodeEmb, for graph representation learning.
• Incorporated a dynamic node-embedding algorithm using a three-layer graph neural network and Given's angle method.
• Proposed an extension, TempNodeEmb++, incorporating time encoding for improved performance.

Main Results:

• Validated TempNodeEmb against seven state-of-the-art benchmark models on diverse real-world datasets.
• Evaluated performance on dynamic protein-protein interaction networks, email networks, and human contact datasets.
• Demonstrated that the proposed models (TempNodeEmb and TempNodeEmb++) outperform existing methods in most cases.

Conclusions:

• The proposed TempNodeEmb algorithm effectively predicts temporal patterns in dynamic networks.
• Graph representation learning offers a computationally efficient approach to link prediction in complex systems.
• TempNodeEmb provides a significant advancement in modeling and predicting the evolution of temporal networks.