Behavioral targeting with social regularization

Behavioral targeting (BT) is a valuable tool for online advertising. In this paper, we study a new problem of incorporating social information into traditional behavior targeting models. Specifically, we present a social regularization based Poisson regression framework for behavior targeting. Based on the observation that social information can be diverse and competing, we furthermore present two specific social regularization terms: the average-based social regularization term and the individual-based social regularization term. To validate the effectiveness of the proposed models, we use the KDDCUP´12 behavior targeting data, issued by the Tecent company in China, as the test bed. The results demonstrate that the proposed models, by incorporating additional social network information, can achieve at least 5% improvement compared to the traditional Poisson regression based model from the CTR lift viewpoint, especially when the historical behavior data is sparse and insufficient.