CHINESE TEXTUAL CLASSIFICATION BASED ON MULTILAYER FEED FORWARD WITH BACK PROPAGATION NEURAL NETWORK

Abstract

Due to the improvement of network technologies, development of text classification methods becomes urgent so as to provide an efficient retrieval in an increasing growth of electronic documents. In this paper, textual classifications based on a supervised neural network are investigated. Since the performance of a neural-based classifier is affected with selection of appropriate textual descriptors, two extraction methods are proposed namely, descriptor extending process and appending process. Meanwhile a parallel classifier is implemented in order to deal with overfitting problem. The performance of proposed models are verified with real textual data. Experimental results show that the parallel model is superior to simple multi-layer feed-forward with back propagation model and the model proposed by Kwok in terms of classification accuracy. Besides, both extending and appending processes indeed improve classification accuracy and speedup training time when they are implemented to different network classifiers.