Refined Spectral Clustering via Embedded Label Propagation

摘要

Spectral clustering is a key research topic in the field of machine learningrnand data mining. Most of the existing spectral clustering algorithms arernbuilt on gaussian Laplacian matrices, which is sensitive to parameters.rnWe propose a novel parameter-free distance-consistent locally linear embedding.rnThe proposed distance-consistent LLE can promise that edgesrnbetween closer data points are heavier.We also propose a novel improvedrnspectral clustering via embedded label propagation. Our algorithm isrnbuilt on two advancements of the state of the art. First is label propagation,rnwhich propagates a node’s labels to neighboring nodes according torntheir proximity.We perform standard spectral clustering on original datarnand assign each cluster with k-nearest data points and then we propagaternlabels through dense unlabeled data regions. Second is manifold learning,rnwhich has been widely used for its capacity to leverage the manifold structure of data points. Extensive experiments on various data sets validaternthe superiority of the proposed algorithm compared to state-of-theartrnspectral algorithms.