The Research Of Multiclass Categorization Algorithm Based On Multiconlitron

With the rapid development of Internet,computer technology and the popularity of smart phones,people's life has becoming more digital and intelligent.At the same time,all kinds of text or log data is also growing explosively.How to quickly classify these data and extract useful information from them which becomes an increasingly important issue.Automatic categorization technology has become a hot research topic in the field of machine learning,because it can greatly shorten the query time,reduce the manpower and resources,improve work efficiency.For the training set with more samples and fewer classes,the paper puts forward a multi-class categorization algorithm based on 1-a-1 method and Multiconlitron.For overall,there is N categories of the training sample.The combination of two categories is for training classifier.Each classifier votes the sample to be classified.The label of the sample ultimately determine the votes.The classification Experiments are done on Reuters 21578,and the experimental results show that compared with multiclass support vector machines 1-a-15,while ensuring the classification accuracy and training speed,the proposed algorithm improves the classification speed greatly.For the training set with fewer samples and more classes,the paper presented a classification algorithm based on multiconlitron and 1-a-r method.1-a-r method is used to convert a multiclass categorization problem to several binary problems.Multiconlitron is constructed for each binary problem in input space.For the text to be classified,its class is decided by multiconlitrons.The classification experiments are done on the Reuters 21578.The experimental results indicate that the proposed algorithm has better classification performance compare with 1-a-r SVMs.It improves the classification precision and classification speed.For the reason of undivided areas on 1-a-1 method and 1-a-r method,a multi-class text classification algorithm based on Multiconlitron and Hadamard ECOC is proposed.Hadamard ECOC is used to convert the multi-class classification problem into a series of binary-class problems.For each of the binary-class problems,the binary classifier is constructed by using Multiconlitron.Hamming distance is used to determine the text category.The classification experiments are done on the Reuters 21578.The experimental results show that,compared with 1-a-r SVMs,1-a-1 SVMs and DAGSVM,the proposed algorithm increases the speed of classification and improve the precision of classification.