Maximizing the effectiveness of advertising campaigns on Twitter

Influence Maximization (IM) has received lots of attention from academia as well as industry. In this work, we study a specific influence maximization problem, selecting a set of seed users to maximize the effectiveness of advertising campaigns on Twitter. This problem requires that the information diffusion model must have the capability to support: (a) an active user can make multiple attempts to activate his neighbors, and (b) a user can accept an advertising message many times. Considering the characteristics of advertisement propagation on Twitter, we propose a diffusion model that inherits the classic independent cascade model and removes the constraints such as single attempt in activating neighbors and taking single influence from other users. The influence probability is calculated based on users' action history including tweet, favorite, reply and retweet. Influence decay is also introduced to reflect the temporal features associated with influence. A new metric 'advertising effectiveness' is defined as the maximization objective. Several existing seed selection algorithms are analyzed based on the proposed diffusion model against a real dataset from Twitter including 3,292 users in Darwin city in Australia. Experimental results and analysis are provided to show the soundness of the proposed model.