Active Semi-Supervised Clustering Based on Multi-View Learning

This paper proposes two new semi-supervised clustering methods based on the combination of multiview,active and semi-supervised learning. Farthest-first traversal scheme is proposed to select the seed set for each cluster. Under the multi-view framework,these two proposed algorithms explore the active learning from two aspects, that is, active seed set selection and active query construction. Experimental results on both Chinese and English data sets show that our proposed algorithms outperform the baseline Constrained KMeans(CKM) and its active version(ACKM).