Research On The Internet Advertisement Click-Through-rate Prediction Method Based On Deep Learning

Internet advertising has continued growth in recent ten years,along with the rapid development of Internet advertisement,compute advertisement got rapid development.In compute advertisement,click-through rate affects the precision of promoting advertisement,The more accurate the click-through rate is,the more the advertisement will satisfy the user's needs,and the more the revenue will be maximized for the advertiser and the media platform.Feature engineering is the key factor to improve the performance of model,One of the important ways to extract the information from the feature is constructing the combination feature method,and the traditional method relies on the artificial experience,For example,the age of the person and the attributes of the advertisement make a combination feature to show that user has different age prefer for different advertising,This method cannot construct the recessive combination feature of more than two dimensions.Therefore,in this paper,research on how to extract the implicit information to improve the accuracy of the prediction model,This paper proposes a method to construct effective combination features using ensemble learning,method,and than,use deep learning to extract the latent feature.The main work of this paper is as follows:1)Proposing use ensemble learning to construct combination feature.Traditional methods of constructing combination feature rely on artificial exper-ience,this methods cannot efficiently extract effective combination feature,this paper propose a method to construct effective combination features by using ensemble learning method,some features are important to the task in the original feature,and the other features are not important,according to the splitting conditions of the feature,the model can form paths of feature combination,and the feature selection is also completed in this process.In this paper,the combination features are constructed by these paths.2)Proposing use deep learning to extract the latent feature.The deep learning has the excellent performance of feature extract,using combination feature made by ensemble learning method to train deep learning model and than to predict click-through-rate.Experiments in AVAZU's click-through rate dataset show that This method can effectively improve the accuracy of advertising rate estimation.