A Multi-Label Classification Algorithm Based on Random Walk Model

Abstract

There are extensive literatures related to traditional single-class and multi-class classification problems,in which each data point is assigned to one category.But in many applications,a data point may belong to more than one category.This kind of problem is called the Multi-Label Classification(MLC) problem.Due to the existing of label relevance,the traditional data-mining methods cannot be directly applied to the MLC problems.This paper proposes a novel MLC algorithm based on the random walk model,called Multi-Label Random Walk(MLRW) algorithm.Firstly,a multi-label random walk graph is built on the training set.As an unlabeled data arrives,a multi-label random walk graph system will be built,on which the random walk processing is carried out.After that,a probability distribution among all labels is obtained.At last,a threshold learning algorithm is proposed based on the MLRW algorithm so that the final prediction on each label is presented.Experimental results on actual data set show that the MLRW algorithm provides an effective solution to the MLC problems.