Multi-level Label Propagation Algorithm Based on Data Reconstruction

Due to standard label propagation algorithm does not use the correct posterior probability of each iteration, and the propagation information of labeled data and unlabeled data are not distinguished during the label propagation process, this paper proposes a multi-level label propagation algorithm Based on data reconstruction. It adds the data which is correctly labeled for each iteration into the labeled data by the nearest neighbor rule Based data editing technique named Depuration, reconstructs the labeled data set, classifies the transition probability matrixes of both labeled and unlabeled data according to their importance. Experimental results show that the proposed algorithm is effective on the performance and convergence rate.