Active Learning for Semi-supervised Classification Based on Information Entropy

Traditional classification of supervised learning needs sufficient labeled data. Unfortunately, in practice, the training data are often either too few, expensive to label, or easy to be outdated. Most of supervised machine learning methods led to poor performance when working on limited tagged data. In recently years, some researches successfully use unlabeled data to help classification. This paper investigated a novel semi-supervised learning method based on active learning with information entropy. An optimization strategy of selecting training instances, based on active learning, was presented. The experiment results show that our method could achieve high performance on small tagged data.