D~2Kernel K-Means Algorithm Applied Research In Social Tagging System

With the increasing requirement of users’personalization in Internet and the spread of communitized Lifestyle, the social tagging system in the web2.0gets a rapid development. Scholars have studied social tagging system from different aspects to take better advantage of social tagging system:the nature of social tagging systems, visualization of the tags, the application of social tagging system and the problems of the social tagging system. This paper mainly focused on the communities of users’interest which is the core of personalized recommendation in social tagging systems. Specific in the following aspects:1At first this article learns about the status of communities of user’s interest in the current research of both domestic and overseas, and introduces the clustering, social tagging system, the definition of the interest communities and the main technical which are used in the community of interest:spectral clustering, hierarchy clustering and K-means algorithm briefly.2Introduced the kernel K-means algorithm. Analyzing the shortcomings of the K-means algorithm, introduce the objective function theorem with the change of the cluster center in K-means algorithm, normal distribution, the characteristics of kernel K-means and D2theory in order to propose an improved effective kernel K-means algorithm. At last it is verified that the proposed algorithm have better performance and prediction accuracy than the kernel K-means by the experiments on UCI datasets3Apply the improved kernel K-means of the user interest communities in social tagging system. Analyzing the shortcomings of the K-means in the user communities of interest, propose semantic similarity and time similarity between users and tags to solve the effect that lack of semantic information and the neglect of time to the performance of clustering while using the algorithm to cluster users. Transform user communities of interest into relevant clustering problem, and propose the new clustering framework, then analyze user communities of interest under the new clustering framework and improved the performance of user clustering.