| With the rapid development of Internet and the burst amount of theinformation, the original manual basis can’t sort out so much text messageseffectively. Using computational techniques to realize automatic textclassification can improve the efficiency of text information management andsave a lot of manpower and material resources, so the text classification researchhas been paid the widespread attention. it is a very meaningful research subject.As a kind of traditional text classification algorithm, centrid classificationhas the advantages of high accuracy and programming is simple and easy toimplement. It is widely used in the actual text classification systems. In order tofurther improve the performance of centrid classification, the professionalscholars in recent years have raised many improvements of the centridclassification. Most of the centrid classification is to change the category ofcomputing into optimization problem, and then to use the optimizationtechnology to solve the problem. Although the type of algorithm improves thecentrid classification performance, it increases running time of the algorithm.As a kind of traditional text classification algorithm, centrid classification has the advantages of high accuracy and programming is simple and easy toimplement. In order to further improve the performance of centrid classification,the professional scholars in recent years have raised many improvements of thecentrid classification. But most of the centrid classification is to change thecategory of computing into optimization problem, and then to use theoptimization technology to solve the problem. Although the type of algorithmimproves the centrid classification performance, it increases running time of thealgorithm.Aiming at the disadvantages of the long running time, this article willintroduce the ideas of dimension reduction method to the text centridclassification algorithms, and put forward the centrid classification algorithmbased on dimension reduction in order to gain the improvement both inperformance and efficiency. The main tasks of the paper are as follows:1. In this paper, first of all, the text classification research status at home andabroad were summarized. It analyzed the method of text representationbased on vector space model and its effect on classification results, andmeanwhile discusses the necessity of the dimension of feature vector spaceand the basic ideas. 2. In order to solve the problem of the long running time caused by the manyfeatures, this paper will propose the centrid classification algorithmPCA-CC and LDA-CC which are based on the linear dimension reduction,LLE-CC, ISOMAP-CC based on the center of the manifolddimensionality reduction text categorization algorithm and LSI-CC basedon semantic dimension reduction. Among them, the linear dimensionreduction method is used principal component analysis and lineardiscriminant analysis. As for the manifold dimensionality reductionmethods, this article uses the linear insertion which is one of the mostrepresentative in the local use of the manifold dimensionality reductionmethod and Laplace feature map manifold dimensionality reductionmethod. However, the dimension reduction based on the semantic methodchooses the method of latent semantic indexing and analyzes their function.In terms of the characteristics of each method, it is embedded in the processof centrid classification. Theoretically, it realizes the centrid classificationbased on the method of linear dimension reduction.3.Carry on the experimental measurements of LDA-CC, the LLE-CC,ISOMAP-CC and LSI-CC algorithm proposed on two standard data set, and compare the effects of classification algorithm from each data set onthe center of dimension reduction with different dimension in detail. Theexperiments shows that this algorithm can not only increase the executionspeed of text classification algorithm but also improve the classificationperformance and reduce memory consumption effectively. |
