A bipartite graph model is brought into the cluster ensemble problem. The object set and hyperedge set are modeled simultaneously via a bipartite graph formulation considering the similarity among instances and the information provided by hyperedges collectively. A normalized spectral clustering algorithm is proposed to solve the bipartite graph partitioning problem, and the final clustering result is attained by performing K-means++ algorithm to partition object set embedded in low dimensional space. Experimental results on several baseline datasets show that the proposed approach is not only well-performed but also high efficient.