For information distortion of cluster fusion algorithm, a new cluster fusion algorithm is proposed, which gives
consideration to both the cluster quality and the diversity of cluster members, and adopts a new method of similarity
measurement. According to the measure result, cluster members are edited, subgrouped and selected, finally a new weighting
function is designed based on the contribution of each cluster member to each category. The practice results show that the
algorithm has better stability and precise compared with other methods.