CA-GNN: A Competence-Aware Graph Neural Network for Semi-Supervised Learning on Streaming Data.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Semi-supervised learning (SSL) is crucial for streaming data mining due to limited labeled examples.
• Graph-based SSL algorithms leverage node interactivity but struggle with unreliable labels and static graph assumptions in dynamic environments.
• Existing methods face challenges in adapting to the evolving nature of streaming data and the potential inaccuracies in graph structures.

Purpose of the Study:

• To address the limitations of traditional graph-based SSL algorithms in dynamic streaming data environments.
• To propose a novel approach, competence-aware graph neural network (CA-GNN), capable of handling unreliable labels and dynamic graph structures.
• To enhance the accuracy and robustness of SSL for streaming data analysis.

Main Methods:

• Developed a competence model within CA-GNN to assess the reliability of individual data points and identify latent semantic correlations.
• Implemented a streaming learning strategy to dynamically update CA-GNN parameters, adapting to evolving graph sequences.
• CA-GNN avoids direct reliance on potentially mislabeled graph information, focusing on data competence and adaptability.

Main Results:

• Experimental results on seven real-world and four synthetic datasets demonstrate superior performance of CA-GNN.
• CA-GNN effectively classifies streaming data, outperforming current state-of-the-art (SOTA) methods in various scenarios.
• The competence-aware approach and dynamic learning strategy significantly improve SSL performance on streaming data.

Conclusions:

• CA-GNN offers a robust and effective solution for semi-supervised learning on streaming data, overcoming key limitations of prior methods.
• The proposed model shows significant potential for applications requiring continuous learning from dynamic and potentially noisy data streams.
• This work advances the field of graph-based SSL by introducing adaptability and reliability assessment for streaming environments.