A+TEXT+CLASSIFIER+BASED+ON+BIOMIMETIC+PATTERN+RECOGNITION.pdf

Proceedings of the Eighth International Conference oll Machine Learning and Cybernetics，Baoding，12—15 July 2009 A TEXT CLASSlFIER BASED oN BIoMIMETIC PATTERN RECoGNITIoN JI-BIN ZHANG，SHUAI CONG，ZHI—MINGXU，QI-SHU PAN School ofComputer science and Technology,Harbin Institute of Technology,Harbin 1 5000 1，China E-MAIL：zjbxgg@hited-cn，congshuai@hit．edu．en，xu／1n@hit．edu．ca，panqs@hit．edu．cn Abstract： in this paper a novel text classification method based on biomimetic pattern recognition(BPR)is proposed．And we implement a BPR text classifier with multi．weight neural network；the model of three．weight neuraI network and i忸 construction method are described in detail．Some text classification experiments have been done to test the performance of onr BPR text classifier；experimental results show that three-weight neuraI network achieved good performance in the text classification task． Keywords： Text Classification；Biomimetic Pattern Recognition； Multi-weight Neural Network； 1． Introduction Text classification aims to map documents to text categories according to their content．At present，many pattern recognition methods have been used in text classification，such as Nafve Bayes【1】，KNN[2】，SVM and so on．In general traditional pattern recognition methods aim to seek all optimal division in the feature space．The fundamental idea of these models is based on the distinction between one class of samples and limited classes of known samples． Biomimetic paRern recognition was firstly introduced by Academician Wang Shoujue[3】in 2002 inspired from classification function of human being．This method emphasizes on the connection of the same class’s samples， and the distinction between one class of samples and unlimited classes of unknown samples．it is much closer to the function of human being． The mathematical tool of the BPR is high-dimensional geometry analysis theory,and its implementation method is multi—weight artificiaI neuraI network，which has been used in many areas．such as speech r