一种具有增量学习能力的PU主动学习算法

PU Active Learning Algorithm with Incremental Learning Ability

在正例和无标记样本增量学习中,初始正例样本较少且不同类别正例的反例获取困难,使分类器的分类和泛化能力不强,为解决上述问题,提出一种具有增量学习能力的PU主动学习算法,在使用3个支持向量机进行协同半监督学习的同时,利用基于网格的聚类方法进行无监督学习,当分类与聚类结果不一致时,引入主动学习对无标记样本进行标记。实验结果表明,将该算法应用于Deep Web入口的在线判断和分类能有效提高入口判断的准确性及分类的正确性。

In positive and unlabeled samples of incremental learning, the initial positive samples are small and positive cases of different types of cases are difficult to get, making classifier classification ability and generalization ability weak. A new algorithm called PU Active Learning algorithm with Incremental learning ability（l-PUAL） is presented, which is applied to Deep Web sources on-line judgments and classification. Experimental results show that it can take advantage of online unlabeled samples to improve the accuracy of judgments and classification correctness.