Label Ranking with Semi-Supervised Learning
Wei Wang, Irwin King · International Conference on Intelligent Information Processing · 2010
Label ranking is considered as an efficient approach for object recognition, document classification, recommendation task, which has been widely studied in recent years. It aims to learn a mapping from instances to a ranking list over a finite set of predefined labels. Traditional solutions for label rankings cannot obtain satisfactory results by only utilizing labeled data and ignore large amount of unlabeled data. This paper introduces a novel Semi-Supervised Learning (SSL) framework by exploiting unlabeled data to improve the performance. Under this framework, we also propose a new Information Gain Decision Tree(IGDT) with aims to make full use of latent information and as such raise the efficiency and accuracy. Then we outline our models involving another two algorithms, Instance Based Learning (IBL) and Mallows Model Decision Tree (MMDT) within this framework. Experiment results demonstrate our approaches can obtain a better performance comparing with only applying labeled data.