Multi-label Classification Using Random Walk with Restart
Jinhong Liu, Juan Yang · 2017
Multi-label classification refers to the task of outputting a label set whose size is unknown for each unseen instance. The challenges of using the random walk method are how to construct the random walk graph and make prediction for testing instances. In this paper, we propose a multi-label classification method based on the random walk with restart model, called ML-RWR. It is derived from the popular KNN algorithm mapping the instances to a KNN based random walk graph. Different from prior work which constructs a complex graph and designs a complex predicting process, we aim at simplifying the complexity of the random walk graph and the complexity of predicting process. Experiments on real-world multi-label datasets show that ML-RWR is superior to those of some well-established multi-label learning algorithm.