Classification learning problem on hypergraph is an extension of multi-label classification problem on normal graph, which divides vertices on hypergraph into several classes. In this paper, we focus on the semi-supervised learning framework, 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 and pure component. Furthermore, we derive a generalization performance bound for near-zero-cut partition with optimal parameter λ.