Investigation of BPNN & RBFN in text classification by Active search

Need of automatic text classification increases with the availability of huge amount of text in internet, news, institutes and organization. The proposed work comprised to deal with the major challenge of getting labeled data for training in classifier, since the availability of labeled data requires the involvement of annotator, is expensive and time consuming. A novel semi supervised text classification algorithm is proposed which makes use of web assisted data by Active search, the proposed algorithm investigates results by applying term weighting method (term frequency)tf and (term frequency.relevance frequency)tf.rf on BPNN (Back Propagation Neural Network)and RBFN (Radial Basis Function Network)classifiers and compared on test data and standard data Mini Newsgroup. Experimental results state that Both BPNN and RBF network performance is comparable for test data in the proposed framework, though RBF Network performance is better and more consistent than BPNN on standard mini newsgroup dataset on the basis of Micro averaged F1measure.