Abstract: Semi_supervised clustering makes use of few supervised information in unsupervised clustering to boost the clustering performance.This paper proposes a semi_supervised clustering algorithm based on seeds set and frequent itemset mining，which mines frequent itemsets in the beginning seeds set and the enlarged seeds set for eliminating the noise data and correcting the mislabeled data to improve the quality of seeds set and enhance the performance of clustering.A weighted χ² measure，as a classification rule evaluation measure，is used to label unlabeled data and they are added into the initial seeds set to enlarge the scale.The experimental results show that the proposed approach effectively reduces the noise data，and not only makes the results more correct but also makes the performance of clustering more better.