Evolving network representation learning based on random walks.

Area of Science:

• Network science
• Machine learning
• Data mining

Background:

• Network analysis is crucial for understanding complex systems.
• Current methods struggle with dynamic, evolving networks.
• Static network embeddings become outdated with network changes.

Purpose of the Study:

• To develop a method for learning continuous network representations of evolving networks.
• To address the inefficiencies and comparability issues of re-calculating embeddings for dynamic networks.
• To enable accurate and scalable network mining on real-world, time-varying networks.

Main Methods:

• Proposes EvoNRL, a random-walk based approach for evolving network representations.
• Dynamically updates random walks to reflect network topology changes without bias.
• Introduces an analytical method to optimize the timing of representation updates for accuracy and efficiency.

Main Results:

• EvoNRL effectively learns accurate, low-dimensional representations for evolving networks.
• The method maintains comparability of embeddings across network updates.
• Experimental evaluation demonstrates superior performance against baseline methods.

Conclusions:

• EvoNRL provides an efficient and effective solution for network representation learning on dynamic networks.
• The approach facilitates continuous and accurate network mining tasks on evolving graph data.
• This work advances the analysis of real-world networks that are inherently dynamic.