Research And Application Of Spectral Clustering Ensemble Algorithms Based On Dynamic Tissue-like Membrane System

With the advent of the information age,the amount of data increases sharply and the complexity of data is higher.How to mine useful information from these data is an important research direction of data mining.Clustering research is a main method of data mining.It is an effective way to analyze data and mine effective information.Traditional clustering algorithms,such as K-means algorithm and K-center algorithm,are unable to deal with increasingly complex data.Spectral clustering is an algorithm based on graph theory.In recent years,spectral clustering algorithm has attracted more and more scholars because of its solid theoretical foundation and excellent clustering effect.However,only one clustering algorithm can not deal with all kinds of heterogeneous data sets and problems encountered in the calculation process.The idea of clustering ensemble integrates the base clustering results generated by multiple learners in a certain way to produce new clustering results.It combines the difference and accuracy of learners,the ensemble algorithm has strong generalization ability and better clustering results.Membrane computing is an efficient and novel computational model which imitates the working mechanism of biological cell membranes.It involves two major fields of computational science and life science,so it is a comprehensive subject with a prospect in the future research.Firstly,the research introduces the status of membrane computing,spectral clustering algorithm,clustering ensemble algorithm and the research trend at home and abroad.The structure and rules of three kinds of membrane computing models are represented.Some concepts,such as graph partition criteria,spectral clustering,clustering ensemble algorithm,learner merging strategy are mentioned.In this part,the innovation and difficulties of this paper are referred to.Secondly,two tissue-like P systems,dynamic organizational P system(DTP system for short)and hybrid dynamic tissue-like P system(HDTP system for short),are proposed in this paper.The former can dynamically adjust the structure according to the input data.The latter combines the tissue-like P system with the cell-like P system on the basis of DTP system to make it a suitable structure for clustering ensemble algorithm.At the same time,the Turing computability of DTP system and HDTP system is proved by registers.Then,two clustering algorithms based on tissue-like P-system are proposed.In the third chapter,an improved spectral clustering algorithm based on Dynamic Tissue-like Membrane system(DTP-ISC algorithm)is proposed.The initial point selection method of K-means algorithm used in the basic spectral clustering algorithm is optimized,which makes the K-means algorithm not easy to be affected by the initial clustering center and outliers,and improves the accuracy of the algorithm.The improved algorithm is combined with the tissue-like P system.Experiments show that the improved algorithm is accurate in clustering.In the fourth chapter,ISC algorithm is used as a learner,and Bagging technology and selective integration strategy are used to design an improved spectral clustering selective clustering ensemble algorithm based on Bagging technology,which is called ISCBE algorithm for short.Then this algorithm is combined with Hybrid Dynamic Tissue-like Membrane System(HDTP-ISCBE algorithm)to improve the effectiveness of the algorithm.In this chapter,HDTP-ISCBE algorithm is compared with DTP-ISC algorithm,spectral clustering algorithm and K-means algorithm,which shows better clustering effect on UCI data sets.Finally,the proposed HDTP-ISCBE algorithm is applied to the research of Weibo user and Xiaohongshu user segmentation.The user data of the two planform is crawled by web crawler software separately.After a series of operations such as data cleaning,coding and feature selection,the data is calculated by HDTP-ISCBE algorithm.The running time of the algorithm is used to measure the efficiency of the algorithm,and analyze the characteristics of user type from the practical significance.Based on this,this paper puts forward countermeasures for enterprises to carry out precision marketing for different types of Weibo users.