Classification learning problem on hypergraph is an extension of multi-label classification problem on normalgraph, which divides vertices on hypergraph into several classes. In this paper, we focus on the semi-supervised learningframework, and give theoretic analysis for spectral based hypergraph vertex classification semi-supervised learning algorithm.The generalization bound for such algorithm is determined by using the notations of zero-cut, non-zero-cut andpure component. Furthermore, we derive a generalization performance bound for near-zero-cut partition with optimal parameter λ.
展开▼
