Rel-MOSS: Towards Imbalanced Relational Deep Learning on Relational Databases

Jun Yin, Peng Huo, Bangguo Zhu, Hao Yan, Senzhang Wang, Shirui Pan, Chengqi Zhang · 2026

Read Rel-MOSS: Towards Imbalanced Relational Deep Learning on Relational Databases by Jun Yin, Peng Huo, Bangguo Zhu, Hao Yan, Senzhang Wang, Shirui Pan, Chengqi Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-07916.txt.

Description

Yin and colleagues confront class imbalance in relational deep learning, where data become entity graphs for graph neural networks. Current methods ignore the imbalance, letting majority nodes swamp minority ones. Rel-MOSS starts with a relation-wise gate that scales signals by relation type, so minority-linked information survives. A relation-guided synthesizer then creates new minority cases while keeping each entity's signature. Tests on 12 datasets show gains of up to about 2 percent in accuracy and 4 percent in G-Mean over leading methods and classic fixes.

Catalog record

CREATORJun Yin, Peng Huo, Bangguo Zhu, Hao Yan, Senzhang Wang, Shirui Pan, Chengqi Zhang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62572
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.07916

Browse the full library · About the Tower of Babel