SEAL: Semisupervised Adversarial Active Learning on Attributed Graphs.

Area of Science:

• Graph Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Active learning (AL) is crucial for addressing label sparsity in graph-structured data.
• Existing AL methods for attributed graphs often fail due to naive strategy combination and separated learning processes.

Purpose of the Study:

• To propose a novel SEmisupervised Adversarial active Learning (SEAL) framework for attributed graphs.
• To enhance node classification performance by leveraging deep neural networks and adversarial learning.

Main Methods:

• Developed a SEAL framework with two adversarial components: a graph embedding network and a semisupervised discriminator.
• Utilized a discriminator-generated divergence score as an informativeness measure for active node selection.
• Implemented a closed-loop system where adversarial components mutually reinforce each other.

Main Results:

• SEAL effectively leverages deep neural networks for attributed graph node classification.
• The adversarial approach significantly improves the informativeness measure for node selection.
• Experimental results demonstrate superior performance compared to state-of-the-art baselines.

Conclusions:

• The proposed SEAL framework offers a powerful and effective solution for active learning on attributed graphs.
• SEAL addresses limitations of existing methods by unifying learning and query processes.
• This approach significantly advances the field of graph-based active learning and node classification.