SBSN is a scalable framework for link sign prediction in signed bipartite graphs, combining efficient subgraph sampling, directed message passing, and node feature distillation to achieve high accuracy and fast inference with low memory cost.