Studies On Semi-supervised Clustering Algorithms Based On Entropy And Divergence

With the continuous development of computer technology and the deepening of its applications,cluster analysis has become an important means and approach for data division or division processing.Maximal entropy clustering algorithm as a fuzzy clustering algorithm,since its launch,it has been developed rapidly.Furthermore,based on the improved maximum entropy clustering algorithm,a semi-supervised clustering technique based on pairwise constraints is introduced to improve the accuracy of clustering.The research and improvement of the technology of Semi-supervised clustering based on pairwise constraints can effectively avoid the wasting of data information and resources.The existing semi-supervised clustering(CE-sSC)overcome the drawback of MEC which is that the MEC cannot use pairwise constraints information.However,the entropy terms of penalty term in CE-sSC are twisted together and this would increase the difficulty of choice for penalty term coefficients.In order to overcome this issue,this paper proposes a new class of semi-supervised clustering algorithm(KL-sSC)which is based on the relative entropy(KL divergence),and presents the paired constraint sample information.The relative entropy term(external information)is generalized to the power divergence(PD)family.PD index could be any real number.And when the number of pairwise constraints is small,one can adjust the PD index with aim to choosing some PD-sSC which clustering performance is better than compared algorithms.Experiment result illustrates that PD-sSC has good clustering performance and its choice of penalty term coefficients is much more simple and more effective compared with CE-sSC.Lastly,by summarizing this paper,some issues which need to be focused and the direction of improvement for future research have been put forward.