METHOD FOR POWER ITERATION CLUSTERING BASED DEFLATION

摘要

The present invention relates to a method for power iteration clustering based on deflation, which is capable of indicating accuracy at a level similar to a spectrum clustering method while maintaining quick arithmetic of power iteration clustering by applying a deflation method to a power iteration clustering method. First, a pseudo eigenvector is generated by applying power iteration to an inputted normalized affinity matrix. And a new normalized affinity matrix is generated by applying the deflation method to the generated pseudo eigenvector. Second, a new pseudo eigenvector is generated by applying power iteration to the new normalized affinity matrix, and another new normalized affinity matrix is generated by applying the deflation method to the new pseudo eigenvector. And generated two or more pseudo eigenvectors are clustered using a K-means algorithm. [Reference numerals] (201) Generate a first pseudo eigenvector by applying power iteration to an inputted affinity matrix; (202) Generate a normalized affinity matrix corresponding to the first pseudo eigenvector by applying a deflation method; (203) Generate a new pseudo eigenvector using the generated normalized affinity matrix; (204) Generate a normalized affinity matrix corresponding to the generated pseudo eigenvector; (206) Cluster the generated pseudo eigenvectors using a K-means algorithm; (AA) Start; (BB) No; (CC) Yes; (DD) End