A semi-supervised clustering algorithm based on local scaling graph and label propagation

Abstract

A semi-supervised clustering algorithm is proposed based on local scaling graph and label propagation. The main idea of this algorithm is that those samples locating in a local neighborhood share the same labels and the global labels changing among the graph is sufficiently smooth. The algorithm firstly introduces a local scaling graph to describe neighborhood among all the samples. Then an objective function and a constraint equation are proposed, which stand for the global smoothness of the category labels' changing and the semi-supervised information respectively. Finally, the clustering task can be expressed by a typical quadratic program, whose optimal solution can minimize the overall smoothness of the labels changing and satisfy the constraint. Experimental results of the algorithm on toy data, digit recognition, and text clustering demonstrate the feasibility and efficiency of the proposed algorithm.