Based on analysising the main step of spectral clustering and finding out its cause of sensitive to the initialization,
a method of spectral co-clustering documents and words based on fuzzy K-harmonic means is proposed. Firstly, the matrix
which is insensitive to the initialization is constructed. Then fuzzy K-harmonic means algorithm is used instead of K-means
algorithm. The experiment result shows that the proposed method not only is initialization insensitive, but also can improve
the accuracy and robustness of clustering results.