Webpages Classification Based on Deep Belief Network Using Images and Text Information

In this paper, Deep Belief Network(DBN) is used for drug-related webpages classification. HTML parsing is used to extract image-label text and body text, FOCARSS method is used to choose effective images. text representation is generated by BOW model, images representation is generated by BOF model. We concatenate images and text representation to generate final representation. It is shown that DBN's classification accuracy is higher than BPNN's classification accuracy, and better than that of single-modal information.