Semi-supervised learning method is a new interesting direction of machine learning approach. It gives the computer a learning ability and makes good use of the obtained knowledge in the application. Semi-supervised learning performs the process of data analysis and mining effectively with the help of few exemplars or little pre-known information. A new Non-Euclidean Space similarity measurement contains the structure information, which is proposed in the Tri-Set computation method. The new similarity measurement not only attentions on the Euclidean Space constraint, but also gives the basic information about the text files. This method is named as Kernel Affinity Propagation (KAP). The method is applied to the benchmark data set Reuters-21578. Further the result is compared with the k-means algorithm and original Affinity Propagation algorithm. The comparison result shows that KAP improves the clustering execution time and provides the better clustering output.

Document Clustering – Semi supervised learning-Smilarity measurement- Message Matrix Computation-Kernel Affinity Propagation method