Densenet121-DNN-Based Hybrid Approach for Advertisement Classification and User Identification

Abstract

Online advertisement plays an important role in human society and is used to influence customers to buy a product or service. But these advertisements are not user-preferred. Sometimes unwanted and malware advertisements appear on the page and make them uncomfortable. Therefore, advertisement classification and displaying the userpreferred advertisement is an important concept in the modern era. For this, in this paper, a hybrid deep learning-based framework is implemented for advertisement classification and user-preferred advertisement. Initially, the Densenet-121-based system is implemented. To extract the features like texts and objects from the input advertisement image, Densenet-121's convolution and pooling layers are used, and then average pooling and softmax layers of Densenet-121 are used for classifying the advertisement into four categories such as automobiles, clothes, foods, and cosmetics. Then, the DNN (deep neural network) technique is implemented to recommend the advertisement based on user preference. Gather the information of the user from user profiles and then process the collected data, if the information is matched with any advertisement then it will show the advertisement as recommended to the user. To evaluate the performance of the proposed approach manually collected data will be used with the accuracy, F1-score, recall, and precision metrics. The proposed models achieve a high accuracy of 99.93%. Furthermore, the outcomes show good precision for various advertisement categories.