Click-Through Rate (Ctr) Prediction for Online Advertising

Abstract

Click-Through Rate (CTR) prediction is used to estimate the probability of a user to click on an item in an online advertising. CTR provides insights in supporting effective online advertising that generates huge advertising revenues and also improve users’ satisfaction. Researchers use various features in their proposed CTR prediction models, and they can be generally classified as advertisement feature, user feature and ad-context feature. Advertisement feature explains “How” an ad serving is, user feature illustrates “Who” is browsing the web pages and lastly the ad-context feature describes “What” the product is, “Where” will an ad displayed and the “Mood” of the web contents from users’ comments. In recent years, there is a transformation on CTR models from the traditional methodologies into deep CTR models. Different studies have emphasized certain features and there is no consensus on the features used in the CTR prediction models to achieve better performance. This paper examines the features used by various CTR prediction models in the online advertising and the various opportunities that are available for future research. Keywords: Ad-context, Click-Through Rate, Deep Interest Evolution Network, Deep Interest Network, Deep Session Interest Network