Feature learning and network structure from noisy node activity data

Area of Science:

• Network Science
• Machine Learning
• Data Mining

Background:

• Traditional network reconstruction methods rely on edge information, which is often unavailable.
• Existing approaches are unsuitable for noisy node activity data with missing values.

Purpose of the Study:

• To present an unsupervised learning framework for network construction using only noisy node activity data.
• To develop a method for learning node vectors and identifying synergistic node roles.

Main Methods:

• Generating random node sequences from node activity data.
• Utilizing a three-layer neural network to train node sequences and obtain node vectors.
• Employing an entropy-based approach for selecting meaningful node neighbors.

Main Results:

• Successfully constructed networks from noisy node activity data.
• Identified nodes with synergistic roles within the constructed networks.
• Demonstrated the framework's effectiveness on both synthetic and real-world datasets.

Conclusions:

• The proposed unsupervised learning framework provides a novel approach for network construction from limited and noisy data.
• The method effectively captures complex node relationships and synergistic roles.
• This framework offers a viable solution for network analysis when only node activity data is available.