基于半监督学习的短文本分类方法

随着万维网的快速普及和发展,Web上出现了大量短文本,如科技文献摘要、微博和电子邮件等.短文本内容短小,相互联系,已标注数据获得困难,导致传统分类方法很难取得较高的分类精度.为了解决短文本分类问题,提出了一种基于半监督学习的迭代分类算法(SS-ICA).它使用较少的已标记数据,利用短文本间的关系进行迭代分类.通过与常用分类方法进行对比表明,在标注数据较少的情况下SS-ICA比其他分类器有更高的分类精度.

Short text classification based on semi-supervised learning

With the rapid development of world wide web,there are more and more short texts emerging on the Web,such as abstract of paper,twitter and email.They are short,keeping links with each other,and there are only a small set of labeled instances available.For the sake of classifying the short text,we present a new method named semi-supervised learning-based iterative classification algorithm(SS-ICA),which has the ability to classify the instances with a small set of labeled instances iteratively.Experiment indicates that SS-ICA significantly increases accuracy when compared to other traditional methods on small training set.