Abstract:In order to solve the problem that K-means algorithms is highly sensitive to initial clusters centers in text clustering, an initial cluster center selection algorithm based on the correlative degree of features was proposed. Features with high correlative degree were chosen after reducing dimensions and a new dataset was created. Subsequently, a candidate initial cluster center set was constructed by merging the similar documents in the new dataset using “OR operation”. Finally, the best centers from the candidate dataset were obtained through computing document density and following the minimax principle. The results of five experimental datasets showed that most F-scores are more than 90%, and entropies are below 0.5. Comparison with the K-means algorithms of Mahout showed that the improved algorithm can choose higher quality centers and produce better clustering results.