| With the development of the Internet, Internet advertising is rapidly changedform, from the traditional text ads, to the image ads, flash ads and video adscurrently, presented in the form of diverse and complex trends. More importantly,Internet advertising has become an important way of profit for many networkservice providers.Classification of Internet advertising, this subject is of great practical value.One hand in real life, company hava an macro understanding of on their ownindividual advertising situation, of course, which could including the collection ofcompetitors advertising. Another hand, we can observe the various sectors of theoverall advertising situation, and thereby forecasting the future development ofeconomic.This article is from the perspective of multi-feature information for classifyingInternet advertising. Mainly to complete the following tasks:(1) According to the characteristics of Internet advertising, summed up many ofits features, including the characteristics of the ad itself, where the web advertisingfeatures, and ad links to a web page features three parts, filter the features base onmissing data rate and the accuracy of classification results.(2) For the choosed features, using different classification methods, get everyfeature the most suitable classification algorithms. The text using SVM, KNN, NBmethod and the best results reach73.80%(the title of linked web page with SVM).Fingerprint image using gray-dimensional histogram approach, the best result reach31.51%(gray-dimensional histogram). Use this method to get the most appropriateclassification algorithm features optimal classification result.(3) Proposing a variety of feature fusion ideas, testing a variety of integrationmethods. Such as: the probability of correction fusion method, Bayesian integrationmethods. Experimental results show that bayesian fusion method improved theclassification results of Internet advertising. Achieve a better classification resultsfrom the initial76.85%up to80.77%. |
PDF Full Text Request