Towards multi-order homophily-aware propagation in multi-view learning

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Multi-view learning integrates diverse data perspectives for enhanced performance.
• Graph neural networks (GNNs) in multi-view learning often overlook view quality and graph heterophily, leading to noise propagation.
• Sub-optimal performance arises from treating all views equally and ignoring similarity matrix heterophily.

Purpose of the Study:

• To propose a multi-order homophily-aware multi-view learning method.
• To mitigate the propagation of heterophily information within graph structures.
• To improve the performance of multi-view integration by addressing noise and view quality.

Main Methods:

• Developed a multi-order decoupling module to encode neighborhood and self-embedding independently.
• Introduced a homophily-aware propagation module to construct a homophily-aware matrix.
• Calculated homophily contribution across different neighborhood orders as propagation confidence for guided aggregation.

Main Results:

• The proposed method effectively mitigates heterophily information propagation.
• Experimental results show superior performance compared to existing state-of-the-art models.
• The approach enhances multi-view integration by intelligently aggregating node information.

Conclusions:

• The multi-order homophily-aware multi-view learning approach offers significant improvements.
• Addressing view quality and graph structure is crucial for effective multi-view learning.
• The proposed method provides a robust solution for integrating information from multiple data perspectives.