A Study On M3-kNN Network And Application In Text Categorization

Today, information technologies are moving fast forward, machine learning and pattern recongnization in computer science are more and more mature and widely used in many areas.Natural language processing is also one of the most important research area. Automatic categorization was developed because information search requirements arise. With the popularity of the Internet, text categorization was widely used in information retrieval; information access; information filtering; information storage and management.This paper make a study on K-Nearest-Neighbour used in Min-Max-Modular neural network, which was called M3-kNN method. The research is focused on voting methods used in M3-kNN to impove the accuracy of text categorization.The work done in this paper includes: Under the context of statistical natural language processing, do study on the Min-Max-Modular neural network. Based on the study on Min-Max-Modular(M3) network, make a futher research on k-NN worked in M3 network, which was called M3-kNN method. The research focus on improving voting method used in M3-kNN method, we design four new voting methods based on voting methods which are originally used in traditional k-NN method. The four new voting methods were tested on news corpus, and we collect the results to make some summary about the relationship between distance and the variation of weight used to determine the category.The significance of this paper is researching the voting method applied in k-NN method which was used in M3 network, especially on improving the accuracy of text categorization via designing new voting methods which are based on traditional voting methods often used in k-NN. This paper also metioned Patent Categorization which is a typical application of text categorization. It do some preparation for further improvement on actual application of patent text categorization.