Signed-Graph Recommendation as Structural Consistency Maximization

This paper argues that existing signed social recommendation systems suffer from inconsistency across structural, propagation, and semantic layers, especially under sparse or noisy data. It proposes SSC-Loop, a unified framework that maximizes structural consistency through three components: ESA-DA for structural consistency, P/N/O propagation for propagation consistency, and a contrastive objective for semantic consistency. Experiments on Epinions target explicit signed social rating prediction, while auxiliary results on Slashdot use a derived link-existence task. The authors state that code is available on GitHub.