Learning to predict ad clicks based on boosted collaborative filtering

This paper addresses the topic of social advertising, which refers to the allocation of ads based on individual user social information and behaviors. As social network services (e.g., Facebook and Morgenstern) are becoming the main platform for social activities, more than 20% of online advertisements appear on social network sites. The allocation of advertisements based on both individual information and social relationships is becoming ever more important. In this study, we first propose the notion of social filtering and compare it with content-based filtering and collaborative filtering for advertisement allocation in a social network. Second, we apply content-boosted and social-boosted methods to enhance existing collaborating filtering models. Finally, an effective learning-based framework is proposed to combine filtering models to improve social advertising. The experiments are conducted based on datasets collected from a social finance web site called Morgenstern. We performed a series of comparison experiments between filtering approaches. The experimental results indicate that the learning-based framework is able to achieve better performance results than fundamental filtering and boosted filtering mechanisms alone.