Research On Semi-supervise Clustering Algorithm And Its Application

Clustering is a basis activity for human being to understand the world as well as an important tool to deal with data in our everyday life. It has wide applications in a lot of fields. In this article we focus on clustering analysis in the semi-supervise clustering algorithm and propose a new method of clustering in semi-supervise based on a carefully analysis of existing methods in literature.In this paper, first has done the thorough research to the cluster analysis, introduced the cluster development and the research present situation in detail, and in this foundation have carried on the summary, the comparison of cluster method possess of the different significance;The information theory correlation knowledge and concepts such as data distribution and optimized theory that has the close relation with the paper research was defined.Then with emphasis has discussed cluster algorithm that Clustering Based on Conditional Distributions in an Auxiliary space.and analyzed this algorithm related nature;Then the maximum-entropy approach is introduced to semi-supervised clustering, and a novel semi-supervised clustering algorithm AMESC based on auxiliary space, maximum entropy and simulated annealing is proposed. A new cluster algorithm is given. This algorithm realizes the efficient clustering by minimizing the cost function iteratively. Our experimental results demonstrate its validity and superiority.In general, unsupervised clustering is only based on primary space. If auxiliary space is considered to incorporate with primary space, unsupervised clustering will become semi-supervised clustering. In this paper, auxiliary space combined with primary space is introduced to the design of the corresponding cost function. And accordingly a novel semi-supervised clustering algorithm APMSC based on both auxiliary space and primary space is proposed. This algorithm realizes the efficient clustering by minimizing the corresponding cost function iteratively.Finally, this algorithm is designed and implemented. Through Our experimental results demonstrate its validity and superiority.