Text Classification by Markov Random Walks with Reward

摘要

We propose a novel model for semisupervised classification by bringing in reward in Markov random walks. Both angle and distance metrics for vectors are combined in this model. Taking advantage of absorbing states, transient analysis of Markov random walks can be performed more easily. Diffusion of unlabeled data points makes our approach suffer less from error propagation for the classification process. The experiment results show that Markov random walks with reward can be efficiently applied in the semi-supervised classification.