Semi-supervised multi-label classification problem is usually decomposed into a set of single-label semi-supervised binary classification problems. However, it results in the ignorance of the inner relationship between labels. A semi-supervised multi-label classification algorithm was presented, which avoided multiple single-label semi-supervised binary classification problems but adopted the overall approach in this paper. On the basis of undirected graph, local learning regularizer for data points and Laplace regularizer for labels were introduced and regularization framework of the problem was constructed. The experimental result shows the proposed algorithm has higher precision and recall.