Research On Key Technologies Of WEB Image And Text Advertisement Recognition

The rapid development of the Internet has promoted the development of Internet advertising industry.Nowadays,advertisements occupy the whole network,and most websites contain some kind of advertisement.Although advertisements can bring benefits to websites or enterprises,it also interferes with users' access to useful information.Therefore,it is of great significance to research on how to identify advertisement on the network.For the image ads and text ads that advertiser use more,most of the existing image advertisement recognition methods are completed by rule matching,which needs to update the rules frequently,and the efficiency of real-time detection needs to be improved.The recognition method based on multidimensional features,feature extraction is more complex.The recognition of text advertisement ignores the data sparseness of short texts,and less consideration of the semantic information of texts,resulting in a low recognition rate.In order to solve the problems of current technology,this paper studies the URL of image advertisement and the semantic information of text advertisement.Firstly,aiming at the low efficiency of rule matching in real-time detection and the complexity of feature extraction based on multi-dimensional features,a real-time traffic oriented web image advertising recognition method is proposed.By studying the link features of image advertisements and combining with the SVM classification algorithm of machine learning,an SVM recognition model based on the link features of image ads is proposed.The accuracy,precision,recall and F-value of the recognition results are compared with the rulebased matching Quero model,the DOM-based multi-dimensional feature recognition method,and the image content-based recognition method.It is verified that the single-dimensional feature using only URL addresses proposed in this paper can improve the efficiency of realtime detection without reducing the accuracy.Secondly,the existing text advertisement recognition methods ignore the sparseness of the data,less consideration of semantic information,and the recognition accuracy needs to be improved.A text advertisement recognition method for short text content is proposed.By studying the semantic information of the text and combining the convolutional neural network,a semantic-based convolutional neural network recognition model is proposed.By comparing the accuracy,precision,recall,and F value with the traditional TD-IDF recognition method and the topic model-based recognition method.It is verified that the proposed semantics of using text can improve the accuracy of the recognition result,and the model as a whole Performance has also improved.