Semi-supervised Sentiment Classification with Random Feature Subspace Method
Guodong Zhou · Zhongwen xinxi xuebao · 2012
Recently,sentiment classification has become a hot research topic in Natural Language Processing.In this paper,we focus on semi-supervised learning paradigm for this task where only small amount of labeled data with many unlabeled samples are available for learning.Specifically,we propose a novel approach to semi-supervised learning for sentiment classification based on random subspace method.First,various random subspaces of the feature space are dynamically generated;Then,co-training algorithm is applied to choose high-confidential samples from the unlabeled data with the subspaces as the different views.Finally,the trained model is updated with the new obtained high-confidential samples.Experimental study across four product domains shows that our approach clearly outperforms the static way of the subspace generation and achieves much better performances than many other existing approaches for semi-supervised sentiment classification.In addition,this paper also explores the issues of different feature subspaces numbers.