| Nowadays,in the Internet era,massive information is contained in a large number of webpages,which has great potential commercial value.Therefore,webpage classification has become an increasingly concerned research field.With the popularization and development of Internet in China,Chinese webpage classification has become an important research topic.Multi-label classification is that there are multiple category labels in the classification system of index data sets,but a sample can be assigned to multiple labels at the same time.Multi-label classification means that a sample can be assigned to multiple tags at the same time.Most of the previous webpage classification methods are based on text classification technology,but the webpage text has its particularity in composition.We think that we can improve the accuracy of webpage multi-label classification by using the related information of external linked webpages.The work of this paper includes the construction of multi-label Chinese webpage dataset,the design of the model based on the current webpage content classification,and the model research of the current webpage classification combined with the external information from link webpages.The core innovation of this paper is to propose the convolutional neural network model of multi-information fusion,which effectively integrates the information of the current webpage and multiple external linked webpages and improves the accuracy of the multi-label classification of the current webpage.In the work of this paper,first of all,a distributed crawler is used to crawl Chinese webpages,and then a simple tagging WEB system is built by ourselves.The Chinese webpages are labeled with multiple tags,and the data set of Chinese webpages with multiple tags is constructed.This paper improves the WOCA-SVM model based on the weight-in-order construction algorithm,and proposes the WOCA-NB model,and modifies the original TextCNN model used for single-label multi-class classification,so that it can be used for multi-tag Chinese webpage classification.These models are classified based on the current web content.More importantly,this paper proposes a webpage classification convolutional neural network model(PageCNN)and two extension models(PageCNN-CLL and PageCNN-WLL)combining the information of external linked webpages,which can effectively integrate text and label information extracted from multiple external linked webpages.In the experimental part,we compare the PageCNN model and its extension models proposed in this paper with WOCA-SVM,WOCA-NB,improved TextCNN model,and the current mainstream multi-label text classification model based on deep learning.The results of experiment show that the PageCNN models are superior to the comparison model in terms of subset accuracy,Hamming loss,macro F1 and micro F1,which verifies that the multi-information fusion method proposed in this paper can effectively process the input information from the current webpage and multiple link webpages,and improve the performance of multi-label Chinese webpage classification. |
