Temporal Network Embedding via Tensor Factorization.

Area of Science:

• Graph representation learning
• Network science
• Machine learning

Background:

• Representation learning on static graphs is impactful.
• Temporal networks, with changing edges, require specialized methods.
• Existing temporal network methods struggle with temporal interdependence.

Purpose of the Study:

• To propose Toffee, a novel approach for temporal network representation learning.
• To effectively encode cross-time information and capture periodic changes in evolving networks.
• To generate improved embeddings for link prediction tasks.

Main Methods:

• Utilizes tensor decomposition for temporal network representation.
• Employs the tensor-tensor product operator to encode cross-time information.
• Focuses on capturing periodic patterns in evolving network structures.

Main Results:

• Toffee successfully encodes temporal interdependence in network data.
• The method captures periodic changes within evolving networks.
• Demonstrates superior performance over existing methods in link prediction.

Conclusions:

• Toffee provides effective embeddings for temporal networks.
• The tensor decomposition approach enhances understanding of dynamic network structures.
• Outperforms prior methods on real-world temporal network datasets for link prediction.