New Rel-MOSS method tackles imbalanced data in relational deep learning

Researchers have introduced Rel-MOSS, a novel approach to address class imbalance in relational deep learning on relational databases. This method aims to prevent minority entities from being overshadowed by majority ones by employing a relation-centric synthetic over-sampling technique. Rel-MOSS utilizes a relation-wise gating controller to modulate neighborhood messages and a relation-guided synthesizer to maintain relational consistency during over-sampling. Experiments show Rel-MOSS improves Balanced Accuracy and G-Mean by up to 2.46% and 4.00% respectively, outperforming state-of-the-art methods. AI

IMPACT Addresses a key challenge in applying deep learning to structured data, potentially improving model performance on imbalanced datasets.

RANK_REASON The cluster contains an academic paper detailing a new method for relational deep learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Balanced Accuracy
• entity classification
• graph neural network
• Jun Yin
• relational databases
• relational deep learning
• Rel-MOSS

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →