An improved K-means text clustering algorithm is proposed. The improvement of the algorithm is based on following two points: 1) selection of initial cluster centers based on distance between text and cluster density

confidence radius is introduced to get cluster density

or the points are selected as initial cluster centers based on the longest distance between text and the maximum of cluster density; 2) text similarity is calculated by Hamming distance based on weight

clustering quality of different algorithm are measured based on silhouettete coefficient.Experimental results show that

compared with the original K-means text clustering algorithm and the algorithm in reference[1]

the proposed algorithm can improve clustering quality.