Application of Machine Learning Algorithms for Display Advertising Effectiveness

For internet-search companies it is critical to show and match relevant ads for the user on the web pages. One of the greatest challenges for advertising platforms is to predict click-through-rate (CTR) of the advertisement to find if it is relevant for users. Most of the companies use machine learning algorithms to improve the quality of the CTR prediction. However, existing methods are based on the resemblance of user query and the ad, ignoring the user interests and reactions of other users to these ads, while other models rely on advertisement click history and are limited when predicting the CTR for new advertisements. There is a growing body of research that recognises the application of neural networks modes for the CTR-prediction based on the query text and advertisement content. However, these models still don’t rely on the advertising response and don’t use the user’s click history. This paper attempts to propose a new supervised machine learning model based on the deep neural networks for the CTR prediction that uses the advertisement information and user click history, to compare this method to some of the existing models, score the model on the dataset of internet-search company and describe a way to integrate this model into the search-company runtime infrastructure.