Research on CTR Prediction for Contextual Advertising Based on Deep Architecture Model

Abstract

Click-Through Rate (CTR) prediction is an important step in internet advertising system because it affects web publisher's profits and advertiser's payment. With the traditional machine learning models having surface architecture, the satisfying results can not be obtained by many prediction methods. This paper proposes a deep architecture model (DBNLR) that integrates deep belief network (DBN) with logistical regression (LR) to deal with the problem of CTR prediction for contextual advertising. In this model, DBN is used for automatically getting abstract and complicated features from original data that consists of contents of advertisements, users' information, click logs and pages information without any artificial intervention and prior knowledge, and then a regression model is adopted to calculate the probability value of CTR prediction. Many experiments on relative datasets show that the DBNLR model, compared with another deep architecture model SAELR, has better value of Area Under Curve (AUC) and improves the effect of CTR prediction for contextual advertising which will produce great economic benefit in the area of internet advertising.