A Novel K-Means Clustering Algorithm Based on Positive Examples and Careful Seeding

Positive and unlabeled learning (PU Learning) is a special semi-supervise learning method. Its most important feature is that training set only includes two parts: positive examples and unlabeled examples. Many real-world classification applications appeal to PU Learning problem. The K-means++ clustering algorithm proposed a new seeding method. This paper describes a semi-supervised learning algorithm for positive and unlabeled examples (PU learning). Our approach extends K-means++, an enhancement to K-means that seeds the algorithm with suitably chosen cluster centers, to such situations. The experiments on the Spam and 20-newsgroup data sets shown that our proposed algorithm has better performances.