Dynamic community detection over evolving networks based on the optimized deep graph infomax.

Area of Science:

• Complex Systems
• Network Science
• Data Mining

Background:

• Dynamic networks exhibit nonlinear features crucial for understanding evolving relationships.
• Current network embedding methods for community detection are computationally expensive due to storing all time-step information.
• Existing methods overlook local network structures, impacting node representation accuracy.

Purpose of the Study:

• To propose a novel optimized dynamic deep graph infomax (ODDGI) method for dynamic community detection.
• To address the limitations of existing methods, including high computational cost and inaccurate node representation.
• To improve the accuracy and stability of community detection in dynamic networks.

Main Methods:

• Utilizes a recurrent neural network (RNN) to capture network dynamism and update parameters efficiently.
• Implements a similarity aggregation strategy to consider node importance, enhancing node representation.
• Applies the ODDGI method to both real-world and synthetic dynamic networks.

Main Results:

• The ODDGI method effectively captures network dynamism without requiring storage of all historical data.
• Improved node representation accuracy through a novel similarity aggregation strategy.
• Demonstrated superior performance compared to state-of-the-art algorithms in clustering accuracy and stability.

Conclusions:

• The proposed ODDGI method offers an efficient and accurate solution for dynamic community detection.
• ODDGI overcomes the limitations of previous methods by integrating RNNs and similarity aggregation.
• Experimental results validate the effectiveness and stability of ODDGI on diverse network datasets.